The Approximate Number System

Before you count, you already know roughly how many. Glance at two handfuls of coins and you can tell which pile is larger without counting either one; the same glance tells you a parking lot is more crowded than it was an hour ago, or that one tree carries more apples than another. This immediate, wordless sense of quantity is produced by the approximate number system, an ancient part of the mind that estimates how many things are present without counting them. It is not precise: it cannot tell 20 from 21, and it grows less reliable as the numbers climb. But it is fast, automatic, and shared far beyond our species, present in newborn infants who have never heard a number word and in animals from monkeys to fish. Its single most telling feature is that its accuracy depends on ratios, not gaps: telling 10 items from 20 is effortless, telling 90 from 100 is hard, even though the difference of ten is identical, because the mind measures number the way it measures brightness or loudness, by proportion. This page traces the approximate number system from that ratio signature through the number-tuned neurons that carry it, its emergence in infancy and its course across the lifespan, the separate system that handles very small quantities, its much-debated link to success in school mathematics, and the sharp modern challenge to whether it is a sense of number at all. Three interactive tools let you find the acuity of a number sense and watch its precision, trace how that acuity changes from infancy to old age, and see for yourself the confound that has thrown the whole field into productive doubt.

The approximate number system (ANS) is a cognitive system for representing the number of objects, sounds, or events in a set rapidly and without counting, yielding estimates that are approximate rather than exact and whose precision is limited by the ratio between the quantities compared (Feigenson, Dehaene, & Spelke, 2004). It is one of the core systems of numerical cognition, evolutionarily ancient and shared across human adults, preverbal infants, and many non-human animals, and it operates over multiple senses, so that the same imprecise magnitude sense applies to seen objects, heard tones, and felt events alike. The importance of the ANS for cognitive psychology is threefold. It is the clearest case of a number that behaves like a perceptual magnitude, obeying Weber's law exactly as brightness and weight do, which places number among the basic dimensions the mind measures by proportion. It has a well-characterized neural substrate in number-tuned neurons of the parietal and prefrontal cortex, making it one of the few high-level cognitive faculties traced to identified cells. And it has become central to the science of mathematical development, because individual differences in the precision of the number sense have been linked, if controversially, to achievement in symbolic mathematics (Halberda, Mazzocco, & Feigenson, 2008; Piazza, 2010). The sections below develop the ratio signature and its measurement, the neural basis, the developmental course, the distinct small-number system, the link to mathematics, and the live controversy over whether the ANS is a sense of number or of magnitude.

A Sense of Number

The approximate number system is what lets a person apprehend quantity at a glance. Shown an array of dots too numerous to count in the time available, people can nonetheless report roughly how many, and reliably choose the larger of two such arrays; this dot-comparison task is the standard laboratory probe of the system (Halberda, Mazzocco, & Feigenson, 2008). The estimates it produces have two defining characteristics. They are approximate, carrying a margin of error that grows with the numbers involved, and they are produced without the verbal counting routine that yields exact answers. Because the system operates before and beneath language, it is available to organisms and individuals with no number words at all, as shown by remote cultures whose language lacks a developed counting routine yet whose members perform approximate arithmetic as accurately as schooled adults (Pica, Lemer, Izard, & Dehaene, 2004), and it delivers the intuitive sense of amount that verbal counting later refines into precision. Theorists group it with a small set of evolutionarily ancient core knowledge systems that structure cognition from early in life, the ANS being the core system dedicated to number (Feigenson, Dehaene, & Spelke, 2004). A crucial property is abstractness: the same system estimates the number of dots in a flash, tones in a sequence, or jumps of a puppet, so it is not tied to any single sense but represents numerosity as such. That abstraction is one reason it is regarded as a genuine number sense rather than a quirk of vision, though, as a later section shows, precisely that claim is now contested.

The Ratio Signature

The behavioral fingerprint of the approximate number system is that discrimination depends on the ratio of two quantities rather than their absolute difference. Choosing the larger of two collections is easy when their ratio is large and hard when it approaches one: 10 against 20 is trivial, 90 against 100 is difficult, though the difference of ten is the same in both. This is exactly Weber's law operating over number, and it means the system does not represent counts precisely but represents each numerosity as a fuzzy magnitude whose imprecision scales with its size. The same signature appears as the numerical distance effect, first reported by Robert Moyer and Thomas Landauer, who found that deciding which of two digits is larger takes longer as the digits grow numerically closer, evidence that even symbolic numbers are compared by an underlying analog magnitude (Moyer & Landauer, 1967). Because discrimination is ratio-limited, the acuity of an individual's number sense can be summarized by a single number, the Weber fraction, written w, which is the smallest ratio that person can reliably tell apart. A small w denotes a sharp number sense that can separate close quantities such as 9 and 10; a large w denotes a coarse one that confuses them. This quantity is what varies across people, across development, and across clinical groups, and it is the workhorse measurement of the entire field. In practice it is estimated not from any single comparison but by fitting the psychometric function to a person's accuracy across many ratios and taking the value of w that best reproduces the observed curve (Halberda & Feigenson, 2008). Figure 1 shows how the Weber fraction determines the psychometric function relating accuracy to ratio, and the demonstration below lets you set an acuity and watch the resulting number sense succeed or fail.

A System in the Brain

The approximate number system is among the few high-level cognitive faculties traced to identified neurons. Recording from monkeys trained to judge whether two displays held the same number of items, Andreas Nieder and colleagues found neurons in prefrontal cortex that were maximally activated by a specific numerosity and largely indifferent to the exact size, shape, or arrangement of the items, so that a cell might fire most strongly for four objects whatever those objects looked like (Nieder, Freedman, & Miller, 2002). These number neurons, later found in parietal cortex as well, are tuned like the analog sense predicts: a cell tuned to four also responds somewhat to three and five, and its tuning grows broader for larger preferred numbers, the neural version of the ratio limit (Nieder & Dehaene, 2009). The human homologue was identified with functional imaging by Manuela Piazza and colleagues, who demonstrated tuning for approximate numerosity in the human intraparietal sulcus, with the same broadening of tuning for larger numbers (Piazza, Izard, Pinel, Le Bihan, & Dehaene, 2004). Dehaene has argued that this population codes number on a compressed, roughly logarithmic internal scale, the neural realization of the Weber-Fechner law for number, which is why equal ratios rather than equal differences feel equally spaced (Dehaene, 2003). The convergence of monkey electrophysiology and human imaging on numerosity-tuned parietal populations is the strongest evidence that the number sense is a specific, biologically grounded system rather than a byproduct of general reasoning. The same parietal territory is implicated in judging other magnitudes, notably space and time, which has led to the proposal that number is one facet of a general magnitude system rather than an isolated faculty (Walsh, 2003).

How Number Is Represented

That the approximate number system obeys the ratio law does not by itself reveal what its internal code looks like, and here a genuine and unresolved debate divides the field. On one account, associated with Dehaene and with Nieder, number is written on a compressed, roughly logarithmic scale: each numerosity is represented with about the same amount of noise, but the values themselves are packed ever closer together as they grow, so that the same ratio spans the same internal distance at any magnitude (Dehaene, 2003; Nieder & Dehaene, 2009). On a competing account, associated with Charles Gallistel and Rochel Gelman, the scale is linear rather than compressed: each numerosity is represented by a magnitude proportional to it, but with noise that grows in proportion to the magnitude, a property called scalar variability (Gallistel & Gelman, 2000). The two models are hard to tell apart because both predict the ratio signature exactly, since constant noise on a compressed scale and proportional noise on a linear scale yield the same ratio-limited discrimination. This is the numerical counterpart of the unresolved psychophysical dispute between a logarithmic and a power law of sensation, and like that dispute it remains open. The distinction is not merely philosophical: the standard model used to measure acuity, the psychometric function of the previous section and the worked example below, is derived from the linear, scalar-variability account, so the very formula that quantifies the number sense already embodies one side of the debate. The linear account also carries the system beyond mere comparison, because magnitudes proportional to number can be combined arithmetically; representing number as an accumulated quantity makes nonverbal addition and subtraction of approximate quantities a natural operation (Gallistel & Gelman, 1992), and training on such approximate arithmetic has been reported to transfer to symbolic mathematics (Park & Brannon, 2013).

Ancient and Shared

That the same numerical abilities appear across distantly related species is central to the claim that the ANS is old and innate. Trained to respond to displays in ascending numerical order, monkeys can order the numerosities one through nine and spontaneously extend the rule to new, larger numbers they were never trained on, showing genuine numerical ordering rather than memorized associations (Brannon & Terrace, 1998). Crucially, animals show the same psychophysical signatures humans do, above all ratio-dependence, so that a monkey, like a person, finds close numerosities harder to tell apart than distant ones. The list of species with a demonstrable approximate number sense now runs from primates through birds to fish, and the shared ratio signature across this range is what licenses the inference to a common, evolutionarily ancient magnitude system underlying it (Nieder & Dehaene, 2009). This comparative evidence matters for cognitive psychology because it separates the approximate sense of number, which humans share with animals and infants, from the exact symbolic number system, which depends on language and culture and is uniquely human.

Origins and Development

The approximate number system is present at the very start of life and changes in precision across the lifespan. Newborns only hours to days old detect number abstractly across the senses, matching a heard numerosity of sounds to a seen numerosity of objects, which shows the system does not wait on experience or language (Izard, Sann, Spelke, & Streri, 2009). By six months, infants reliably discriminate large sets at a two-to-one ratio, telling eight items from sixteen but not eight from twelve (Xu & Spelke, 2000). From there acuity sharpens steadily through childhood, the Weber fraction shrinking so that older children and adults resolve progressively closer ratios, a trajectory charted from three-year-olds through adulthood (Halberda & Feigenson, 2008). A massive internet-based sample mapped the full arc, showing that number-sense acuity improves through childhood and adolescence, peaks in the early thirties, and then declines gradually into old age (Halberda, Ly, Wilmer, Naiman, & Germine, 2012). Table 1 collects this developmental course. The lifelong presence and lawful maturation of the system, summarized in the demonstration that follows the table, are part of why it is treated as a foundational rather than an acquired capacity.

Table 1
Acuity of the Approximate Number System Across the Lifespan

Note. Acuity is summarized by the Weber fraction, the smallest ratio reliably discriminated, which falls (improves) through childhood and rises (worsens) in old age; specific values vary with task and observer. Sources are, in order, cross-modal newborn matching (Izard, Sann, Spelke, & Streri, 2009), infant looking-time discrimination (Xu & Spelke, 2000), the childhood trajectory (Halberda & Feigenson, 2008), and the lifespan curve (Halberda, Ly, Wilmer, Naiman, & Germine, 2012).

Two Systems for Number

The approximate number system is not the only way the mind handles quantity, and its boundary is a second system that behaves quite differently. Small collections, up to about three or four items, are apprehended by a separate mechanism of parallel individuation, which represents each item as a distinct object and is precise rather than approximate (Feigenson, Dehaene, & Spelke, 2004). The behavioral marker of this second system is subitizing, the instantaneous and errorless report of small quantities: shown one, two, or three dots, people answer immediately and without mistakes, and their response time barely changes across those values, whereas beyond four it climbs steeply as slow counting or ratio-limited estimation takes over. Lana Trick and Zenon Pylyshyn characterized this discontinuity as the signature of a limited-capacity, preattentive stage of vision distinct from the approximate system (Trick & Pylyshyn, 1994). The decisive point for the number sense is that subitizing does not obey the ratio limit at all: precision over the numbers one to four is far higher than any Weber fraction would predict, a direct violation of Weber's law and strong evidence that a dedicated small-number mechanism operates alongside the approximate one (Revkin, Piazza, Izard, Cohen, & Dehaene, 2008). Two systems, then, cover the number line: a precise one for small sets that escapes Weber's law, and the approximate one for everything larger that obeys it, a division treated more fully in the discussion of Weber's law and the two number systems.

The Number Sense and Mathematics

The finding that carried the approximate number system out of the laboratory is its apparent connection to symbolic mathematics. Justin Halberda, Michèle Mazzocco, and Lisa Feigenson reported that the acuity of fourteen-year-olds' number sense correlates with their mathematics achievement scores reaching back to kindergarten, and that this holds independent of general intelligence and other cognitive skills (Halberda, Mazzocco, & Feigenson, 2008). The proposal that grew from this is that the ANS serves as a start-up tool, an evolutionarily given foundation of numerical meaning onto which the symbolic system of exact number is later built (Piazza, 2010). Several strands support a link. Acuity correlates with symbolic math in both children and adults; preschool acuity predicts later math achievement; and, at the clinical extreme, children with developmental dyscalculia, a specific difficulty with mathematics despite adequate schooling, show markedly poorer number-sense acuity than typically achieving peers, with an acuity deficit that tracks the disorder (Piazza et al., 2010; Mazzocco, Feigenson, & Halberda, 2011). The strongest causal test is training: if the number sense underpins symbolic math, sharpening it should transfer, and Jinjing Park and Elizabeth Brannon reported that training on approximate arithmetic improved adults' symbolic math proficiency (Park & Brannon, 2013). Two cautions belong to any honest account. A meta-analysis of the many studies finds the association between nonsymbolic number acuity and mathematical competence real but modest in size (Schneider et al., 2017), and, as the next section describes, the causal interpretation is directly disputed. The link is genuine and important, but its magnitude and mechanism remain open questions.

Is It a Sense of Number?

Two debates define the frontier of ANS research, and a rigorous account must state both. The first concerns the mathematics link. Camilla Gilmore and colleagues found that once inhibitory control is measured, it, rather than number acuity, carries much of the correlation with mathematics achievement, suggesting that the widely cited relationship is partly confounded by domain-general executive function and that filtering irrelevant information, not number acuity as such, may drive part of the association (Gilmore et al., 2013). The second and deeper challenge questions whether the system senses number at all. Any real collection has continuous properties, total surface area, density, contour length, that tend to grow with number, so a task meant to probe number can be solved by attending to those magnitudes instead. Tali Leibovich and colleagues marshal this point into a direct alternative to the standard view: they argue that humans and animals are born to perceive continuous magnitude, not discrete number, and that the number sense is acquired through learning the correlation between magnitude and number, so that what looks like a sense of number is better described as a sense of magnitude (Leibovich, Katzin, Harel, & Henik, 2017). The debate is not settled, and much of it turns on stimulus control: how thoroughly an experiment equalizes the continuous cues determines how confidently a result can be attributed to number rather than area or density. The demonstration below makes that confound visible, showing why controlling for continuous magnitude is the pivotal methodological question in the field, and why the answer bears directly on whether the approximate number system deserves its name.

Worked Example

The standard model of the number sense makes its ratio limit quantitative. The probability of correctly choosing the larger of two sets is governed by the ratio of their difference to a scaled measure of their sizes, with the scaling set by the Weber fraction: accuracy is the normal probability of a quantity equal to the difference divided by the Weber fraction times the square root of the sum of the two squared numerosities (Halberda & Feigenson, 2008). This is the linear, scalar-variability form discussed above; a logarithmic-coding model fits the same data about equally well, so the numbers here illustrate the ratio law rather than adjudicate the underlying scale. Take a sharp adult number sense with a Weber fraction of 0.15. Comparing 10 against 20, a two-to-one ratio, places the difference far out on the favorable side of the curve and yields near-certain accuracy, above 99 percent. Comparing 18 against 20, the same difference of two dots but a ratio of only 1.11, drops accuracy to about 69 percent, because the ratio, not the difference, is what the system represents. Now hold the comparison fixed and coarsen the number sense to a Weber fraction of 0.30, the sort of value seen earlier in development. The easy 10-versus-20 comparison falls from above 99 percent to about 93 percent, and the hard 18-versus-20 comparison from 69 percent to about 60 percent, barely above chance. Two lessons follow directly. Within any one person, difficulty is fixed by the ratio, so the same numerical gap is trivial at a favorable ratio and near-impossible at an unfavorable one. And across people or ages, a smaller Weber fraction, a sharper number sense, lifts accuracy at every ratio, which is exactly why the Weber fraction, and not any single comparison, is the measurement that captures the acuity of the approximate number system.

Why It Matters

The approximate number system matters first as the foundation of numerical thought. It is the intuitive sense of quantity that precedes counting, is shared with infants and animals, and supplies the raw meaning of amount that symbolic mathematics later makes exact, which is why it anchors modern theories of where number comes from (Feigenson, Dehaene, & Spelke, 2004). It matters second as a rare bridge from behavior to biology: the ratio signature seen in a person choosing the larger of two dot arrays is the same signature carried by numerosity-tuned neurons in the parietal cortex, making number one of the few abstract concepts followed from psychophysics down to identified cells (Nieder & Dehaene, 2009). It matters third for education and clinical practice, because the acuity of the number sense is linked, however contested in mechanism, to mathematical achievement and is impaired in dyscalculia, raising the possibility that a basic perceptual capacity could be assessed and perhaps trained to support mathematics learning (Halberda, Mazzocco, & Feigenson, 2008; Park & Brannon, 2013). And it matters as a live scientific problem, because the challenge that the system may sense continuous magnitude rather than number goes to the heart of what it is and reaches back into every experiment that has claimed to measure it (Leibovich, Katzin, Harel, & Henik, 2017). A capacity we share with newborns and fish turns out to sit at the origin of one of the most powerful cultural achievements of the human mind, and understanding it is understanding where mathematics begins.

Key Researchers

Stanislas Dehaene. Professor of Experimental Cognitive Psychology at the Collège de France; a leading figure in numerical cognition who proposed the logarithmic mental number line and, with colleagues, articulated the core-systems account and identified numerosity tuning in the human brain.
Faculty Page · ORCID · Google Scholar · Wikipedia

Elizabeth S. Spelke. Marshall L. Berkman Professor of Psychology at Harvard University and director of the Laboratory for Developmental Studies; her infant studies established that the ratio-limited number sense is present before language, extending even to newborns.
Faculty Page · ORCID · Google Scholar · Wikipedia

Justin Halberda. Professor of Psychological and Brain Sciences at Johns Hopkins University; he showed that individual differences in the numerical Weber fraction correlate with mathematics achievement and charted the acuity of the number sense across childhood and the lifespan.
Faculty Page · ORCID · Google Scholar

Lisa Feigenson. Professor of Psychological and Brain Sciences at Johns Hopkins University and co-director of the Laboratory for Child Development; a central figure in delineating the two core systems of number and the acuity deficit underlying dyscalculia.
Faculty Page · ORCID · Google Scholar · Wikipedia

Elizabeth M. Brannon. Edmund J. and Louise W. Kahn Term Chair in the Natural Sciences at the University of Pennsylvania; her comparative work established shared numerical abilities across monkeys and humans, and her training studies test whether sharpening the number sense aids mathematics.
Faculty Page · Google Scholar · Wikipedia

Manuela Piazza. Professor of Cognitive Neuroscience at the Center for Mind/Brain Sciences, University of Trento; she identified numerosity tuning in the human intraparietal sulcus, charted the acuity deficit in dyscalculia, and framed the number sense as a start-up tool for symbolic mathematics.
Faculty Page · ORCID · Google Scholar

Key Terms

Frequently Asked Questions

What is the approximate number system?
The approximate number system is a cognitive system that estimates how many items are in a set quickly and without counting, giving rough rather than exact answers. It is shared across humans, infants, and many animals, and its precision is limited by the ratio between the quantities being compared (Feigenson, Dehaene, & Spelke, 2004).

How is it different from counting?
Counting is a slow, exact, verbal procedure that yields a precise number, whereas the approximate number system delivers an immediate, wordless estimate with a margin of error. The approximate system works before language and is available to organisms that cannot count, supplying the intuitive sense of amount that counting later makes exact (Halberda, Mazzocco, & Feigenson, 2008).

What is a Weber fraction for number?
The Weber fraction is the smallest ratio of two quantities a person can reliably tell apart, and it summarizes the acuity of that person's number sense. A small Weber fraction means a sharp sense that can separate close quantities such as 9 and 10, while a large one means a coarse sense that confuses them (Moyer & Landauer, 1967).

Do animals have a number sense?
Yes. Many species, from monkeys to birds to fish, can discriminate and order quantities, and they show the same ratio-dependent signature that humans do, which is strong evidence for a shared, evolutionarily ancient system. Monkeys can even order the numerosities one through nine and extend the rule to new numbers (Brannon & Terrace, 1998).

Is the number sense present in babies?
The system is present from birth. Newborns only days old match the number of sounds they hear to the number of objects they see, and by six months infants reliably discriminate large sets that differ by a two-to-one ratio, showing the sense does not depend on language or schooling (Izard, Sann, Spelke, & Streri, 2009).

Does the number sense affect math ability?
Individual differences in number-sense acuity correlate with mathematics achievement, preschool acuity predicts later math, and children with dyscalculia show poorer acuity. The link is real but its size is modest and its causal basis is debated, with some studies attributing part of it to domain-general skills (Halberda, Mazzocco, & Feigenson, 2008).

Can the number sense be trained?
Some studies report that training on approximate number tasks improves symbolic arithmetic, which would suggest a causal role for the number sense in mathematics. The training literature is promising but contested, with debate over the reliability and interpretation of the transfer effects (Park & Brannon, 2013).

Is it really a sense of number?
This is actively disputed. Because continuous properties such as area and density covary with number, some researchers argue that people are really sensing magnitude rather than number, and that the number sense is learned from that correlation, a challenge that turns on how well experiments control the non-numerical cues (Leibovich, Katzin, Harel, & Henik, 2017).

References