Trail Making Test

Abstract

The Trail Making Test is a type of neuropsychological test: a brief, paper-and-pencil measure of visual attention, processing speed, and cognitive flexibility given in two parts. In Part A the examinee connects numbered circles in ascending order as fast as possible; in Part B the line alternates between an ascending number sequence and an ascending letter sequence (1-A-2-B-3-C). Completion time is the score, and the contrast between the parts is its diagnostic value: Part A indexes visual search and graphomotor speed, while the switching demand Part B adds loads working memory and set-shifting. Because raw seconds mean little without context, scores are read against age- and education-stratified norms, and derived indices, the B minus A difference and the B over A ratio, isolate the switching cost. Reitan validated it as an indicator of organic brain damage.

Keywords: Trail Making Test, set-shifting, processing speed

The Trail Making Test (TMT) looks almost trivial — connect the dots — yet its two-part design turns a simple visuomotor task into a sensitive probe of the frontal-lobe functions that decline earliest in aging and injury. Its endurance for over eighty years rests on a single elegant idea: administer nearly the same task twice, changing only one cognitive requirement, and the difference between the two scores localizes that requirement. This article sets out what each part measures, how the derived scores separate switching from speed, why interpretation is impossible without stratified norms, and what neuroimaging and lesion studies reveal about the brain systems the test recruits.

What the Trail Making Test Is

The Trail Making Test is a timed connect-the-circles task with two parts, each preceded by a short practice sample. In Part A, 25 circles numbered 1 through 25 are scattered across a page, and the examinee draws a single continuous line joining them in numerical order — 1 to 2 to 3, and so on — as quickly as possible without lifting the pen. In Part B, the 25 circles contain both numbers (1–13) and letters (A–L), and the line must alternate between the two sequences: 1 to A, A to 2, 2 to B, B to 3, continuing until 13. The examiner times each part to the second and, in the standard procedure, does not stop the clock for errors; instead the examiner points out each error as it occurs so the examinee corrects it, and the correction time — not an error count — is the penalty (Bowie & Harvey, 2006).

The test is a member of the broader family of neuropsychological tests — standardized behavioral tasks whose scores are referred to normative distributions to infer the integrity of specific cognitive systems. Within that family the TMT is prized for its brevity (two to five minutes), its minimal equipment, and its unusual sensitivity to diffuse and frontal damage. It originated as a subtest of the U.S. Army Individual Test Battery in 1944 and entered clinical use when Ward Halstead and Ralph Reitan incorporated it into the Halstead–Reitan Neuropsychological Battery; Reitan's demonstration that it discriminated patients with organic brain damage from controls established its clinical credibility (Reitan, 1958).

Demonstration 1 — Trace the trail

Connect the circles in order as fast as you can. In Part A follow the numbers 1 → 2 → 3. In Part B alternate number and letter: 1 → A → 2 → B. The circles sit in the same places both times, so any extra effort in Part B comes from the switching rule alone.

Part B usually feels slower even though the layout is identical, because you must hold two sequences in mind and switch between them at every step. Computed locally, not stored; the eight-circle layout is a simplified stand-in for the 25-circle clinical form.

What makes the two-part structure powerful is that Parts A and B share almost every surface feature — the same page size, the same 25 circles, the same instruction to connect them in order under time pressure, the same pen and hand. They differ in exactly one respect: Part B requires the examinee to maintain two ordered sequences simultaneously and switch back and forth between them. Any performance gap between the parts can therefore be attributed, in the ideal case, to that added switching demand rather than to differences in perception or motor control (Gaudino, Geisler, & Squires, 1995).

What Each Part Measures

Part A is dominated by visual scanning and graphomotor speed: the examinee must locate the next number in a cluttered field and move the pen there. It loads processing speed and visual attention with only a light demand on higher-order control. Part B retains all of that and adds cognitive flexibility — the executive capacity to hold an alternating rule in working memory, keep track of one's place in two sequences at once, and inhibit the prepotent tendency to keep following a single ascending series (Sánchez-Cubillo et al., 2009).

Componential studies that regress TMT times on independent measures of candidate abilities find that Part A is carried largely by perceptual speed and motor function, whereas Part B additionally recruits working memory and set-shifting; the two parts are not pure, but the added variance in B aligns with executive control (Crowe, 1998). Experimental dissociations point the same way: when the alternating requirement is removed but the visual complexity of B is preserved, completion time drops toward Part A levels, implicating switching rather than mere clutter as the source of B's difficulty (Arbuthnott & Frank, 2000).

There is, however, a longstanding debate about which executive process Part B taxes. Some evidence frames it as cognitive flexibility — the ability to shift set — while other work argues the harder demand is maintaining set: keeping the alternating rule active against interference rather than switching per se (Kortte, Horner, & Windham, 2002). A neurobehavioral decomposition suggests both matter, with distinct components of working memory, inhibition, and visuomotor sequencing each contributing to the score (Misdraji & Gass, 2010).

Demonstration 2 — Difference and ratio scores

Part B contains everything Part A demands plus switching. Set each completion time and watch the derived scores. The bar splits the Part B time into the component shared with Part A (search and motor speed) and the residual that reflects the switching cost.

At A = 40 s and B = 105 s the difference is 65 s and the ratio 2.63 — a normal, proportionate switching cost. Push Part A up while holding B fixed and the ratio collapses toward 1: someone who is simply slow on both parts has little *extra* switching cost, which is what the derived scores are built to reveal. Computed locally, not stored.

Derived Scores: Separating Switching From Speed

Because Part B contains Part A's demands plus switching, the raw B time confounds two things: how fast a person searches and draws, and how costly it is for them to alternate. Two derived indices try to strip out the shared component.

The difference score, B − A, subtracts the time attributable to visuomotor speed and search (estimated by A) from the total B time, leaving a residual meant to reflect the switching cost in the same units (seconds). The ratio score, B / A, does the same by division, expressing B as a multiple of A and thereby normalizing the switching cost to each individual's own baseline speed. A patient who is globally slow — say from motor slowing or low vision — may have an alarming raw B time yet a normal B − A or B / A, revealing that the slowness is peripheral rather than executive (Sánchez-Cubillo et al., 2009).

Neither derived score is a perfect purifier. Subtraction assumes the two parts draw on speed identically and additively, which holds only approximately, and the difference score can be unreliable when A and B are both short. The ratio has better claim to isolating switching in some analyses but is noisier at the low end. Still, reporting a derived index alongside the raw times is standard practice precisely because it guards against mistaking generalized slowing for an executive deficit.

Why Raw Scores Need Norms

A completion time in seconds means nothing on its own. Trail-making performance slows substantially with age and improves with years of education, and the effect is large enough to swamp clinical signal if ignored: a Part B time that is unremarkable for a 75-year-old with eight years of schooling would be a red flag in a 30-year-old graduate. The most widely used interpretive framework stratifies norms by age band and education level, converting a raw time into a percentile or z-score within the examinee's demographic stratum (Tombaugh, 2004).

Later normative studies extended and refined this scheme, adding coverage for older adults and relating TMT scores to other neuropsychological measures so that a clinician can situate a trail-making result within a broader profile (Llinàs-Reglà et al., 2017). The practical rule is unambiguous: never interpret a raw TMT time without referring it to age- and education-appropriate norms.

Demonstration 3 — A raw time only means something against norms

The same number of seconds is normal for one person and alarming for another. Enter a completion time, pick the examinee's age band and education level, and read the standardized score and percentile within that stratum.

The default — Part B, 105 s, age 65–69, 0–12 years of education — lands at z = +0.18, a touch slower than average and well within limits. Hold the time at 105 s and switch the age band to 25–34: the identical performance now sits far out in the slow tail and trips the flag. Norms are illustrative, patterned on Tombaugh (2004), not for clinical use; computed locally, not stored.

Clinical Sensitivity and Error Analysis

The TMT's clinical reach comes from Part B's sensitivity to the frontal and diffuse pathology that characterizes many acquired and degenerative conditions. Slowed Part B times, and disproportionately large B − A differences, appear in traumatic brain injury, cerebrovascular disease, and the earliest stages of neurodegeneration. Beyond completion time, the errors examinees make carry information: sequencing errors (following the wrong series) and set-loss errors rise across the continuum from normal aging to mild cognitive impairment to dementia, and error patterns can distinguish these groups even when times overlap (Ashendorf et al., 2008).

This is why the test is a staple of dementia screening batteries and post-injury assessment: it is quick, it is hard to fake in the ordinary direction, and it stresses exactly the control functions that decline early. Its weakness is specificity — a slow Part B signals that something is wrong with the fronto-executive system but not what, so the TMT flags rather than diagnoses.

The summary table below contrasts the two parts on the dimensions that matter for interpretation.

Worked Example

Consider a 68-year-old with 12 years of education who completes Part A in 40 seconds and Part B in 105 seconds. The two derived scores follow directly:

- Difference score: B − A = 105 − 40 = 65 seconds. This is the time attributable to the switching demand once the shared search-and-draw component (estimated by A) is removed. - Ratio score: B / A = 105 / 40 = 2.625. Part B took just over two and a half times as long as Part A, squarely within the typical two-to-three-fold range, so the switching cost is proportionate rather than exaggerated.

Now situate the raw times in the examinee's demographic stratum. Using illustrative age-65-69, ≥12-year-education norms (Part A: mean 38 s, SD 12; Part B: mean 98 s, SD 40) — the same schematic table the interpretation demo uses, patterned on Tombaugh (2004) and not for clinical use:

- Part A: z = (40 − 38) / 12 = 0.17, roughly the 57th percentile of completion time — essentially average. - Part B: z = (105 − 98) / 40 = 0.18, roughly the 57th percentile — again within normal limits.

The interpretation is coherent: this examinee is a touch slower than the stratum average on both parts, but the ratio of B to A is normal and neither z-score approaches a clinical cutoff. Had the same raw times come from a 30-year-old graduate, both z-scores would be far larger and the profile would warrant concern — which is exactly why the raw seconds cannot be read without the norms.

Discussion

The Trail Making Test endures because its subtraction logic is close to a natural experiment: two nearly identical tasks that differ in one cognitive requirement let the clinician read an executive cost off the gap between them. That logic is also its main limitation. The parts are not perfectly matched — Part B is visually denser and has more stimuli in play — so the difference and ratio scores isolate switching only approximately, and analyses that model TMT performance with latent abilities show that processing speed accounts for a surprisingly large share of the variance even in Part B (Salthouse, 2011). The test measures a blend, and its interpretive art lies in deciding, for a given patient, whether a slow Part B reflects switching failure or simply generalized slowing.

A second theme is the tension between flexibility and set-maintenance accounts of Part B. Both survive because both are partly right: alternating between sequences requires switching set, but it also requires holding the alternation rule stable against the pull of a single ascending series, and different patients fail for different reasons. This is consistent with the broader picture of executive function as a family of separable but correlated control processes rather than a single faculty. The influential unity-and-diversity account formalizes exactly this: a latent-variable analysis of complex frontal-lobe tasks found shifting, updating, and inhibition to be moderately correlated yet clearly separable functions, with trail-making-type tasks loading most heavily on the shifting component (Miyake et al., 2000).

Current Directions

Two active research fronts are reshaping how the test is used. The first is neuroanatomical localization. Lesion-symptom mapping and functional-imaging studies have moved beyond the simple claim that Part B is frontal to identify a distributed network — dorsolateral prefrontal cortex, but also parietal and subcortical white-matter tracts — whose integrity predicts trail-making performance, with dissociable contributions to Part A and Part B (Varjacic et al., 2018). In older adults, Part B time tracks processing speed, cortical thinning, and white-matter microstructure, tying the behavioral score to measurable structural decline (MacPherson et al., 2017).

The second front is digital administration. Tablet and stylus versions capture not just total time but the full movement trajectory — pauses at decision points, hesitations before a switch, pen lifts, and micro-corrections — decomposing the single completion-time score into component sub-processes that the paper test collapses. Multicomponent analyses of digital trail-making show that these process measures carry diagnostic information beyond the total time, promising earlier and more specific detection of executive decline (Fellows et al., 2017). Whether digital norms can be mapped onto the vast paper-and-pencil normative base remains an open methodological question.

Common Misconceptions

'Part B measures executive function and Part A measures nothing important.'

Part A is not a throwaway; it is the baseline that makes Part B interpretable. Without a Part A time one cannot tell whether a slow Part B reflects a switching deficit or ordinary visuomotor slowing, which is exactly why the derived scores subtract A out. The two parts are diagnostic only as a pair (Sánchez-Cubillo et al., 2009).

'The raw completion time is the score.'

A raw time is meaningless until it is referred to age- and education-stratified norms. The same 90-second Part B is normal for one examinee and pathological for another; only the normed percentile or z-score is interpretable (Tombaugh, 2004).

'Errors do not count because the clock is not stopped for them.'

Errors are penalized indirectly, since correcting them adds time, and their pattern is itself informative. Sequencing and set-loss errors help distinguish normal aging from mild cognitive impairment and dementia even when times are similar (Ashendorf et al., 2008).

'A slow Part B diagnoses frontal-lobe damage.'

The test is sensitive but not specific. A poor Part B signals fronto-executive dysfunction of some kind; the underlying network is distributed across prefrontal, parietal, and subcortical regions, so it cannot localize a lesion or name a disease (Varjacic et al., 2018).

Glossary

Cognitive flexibility.

The capacity to shift attention and behavior between tasks, rules, or mental sets in response to changing demands; the executive process Part B is thought to tax by requiring alternation between number and letter sequences.

Difference score (B minus A).

A derived index computed by subtracting Part A time from Part B time, intended to isolate the switching cost by removing the shared visuomotor and search demand.

Executive function.

The family of top-down control processes, including set-shifting, working memory, and inhibition, that coordinate goal-directed behavior; the fronto-executive system Part B is sensitive to.

Graphomotor speed.

The speed of pen-and-hand movement in producing marks on paper; a peripheral contributor to both parts that derived scores attempt to control for.

Halstead-Reitan Neuropsychological Battery.

A standardized set of clinical neuropsychological tests, assembled by Halstead and Reitan, that brought the Trail Making Test into routine assessment of brain dysfunction.

Mild cognitive impairment.

A stage of cognitive decline greater than expected for age but short of dementia; a condition in which Part B slowing and error patterns often appear early.

Percentile.

The percentage of a reference sample scoring at or below a given value; the normed form in which a raw completion time is expressed for interpretation.

Processing speed.

The rate at which elementary cognitive operations are carried out; a dominant contributor to Part A and a substantial one to Part B.

Ratio score (B over A).

A derived index expressing Part B time as a multiple of Part A time, normalizing the switching cost to each individual's own baseline speed.

Sequencing error.

A trail-making mistake in which the examinee connects to the wrong next target in the series; its frequency helps separate normal aging from cognitive impairment.

Set-loss error.

A Part B mistake in which the examinee abandons the alternating rule and reverts to a single sequence, indexing a failure to maintain the active task set.

Set-shifting.

The executive operation of disengaging from one active task set and engaging another; the mechanism most often invoked to explain Part B's added difficulty.

Stratified norms.

Normative reference values partitioned by demographic variables, for the Trail Making Test by age band and education level, against which a raw time is converted to a percentile or z-score.

Visual search.

The scanning process of locating a target among distractors in a cluttered field; the dominant demand of Part A as the examinee finds each next circle.

Working memory.

The system that holds and manipulates information over short intervals; on Part B it keeps both the number and letter sequences active and tracks the examinee's place in each.

z-score.

A standardized value expressing how many standard deviations a raw score lies from its stratum mean; the metric that places a completion time within its demographic norm.

Key Researchers

Christopher R. Bowie. Clinical neuropsychologist at Queen's University; co-author of the widely cited Nature Protocols standard for administering and scoring the TMT. ORCID - Faculty - Scholar

Nele Demeyere. Post-stroke cognition researcher at the University of Oxford; co-author of the review synthesizing the neural substrates of trail-making performance. ORCID - Faculty - Wikidata

Céline R. Gillebert. Cognitive neuroscientist at KU Leuven working on lesion-symptom mapping of attention and executive control; senior author of the TMT neural-signatures review. ORCID - Faculty - Scholar - Wikidata

Ward C. Halstead (1908–1968). Founder of human neuropsychology at the University of Chicago; the Halstead battery, later the Halstead–Reitan battery, brought the TMT into standard clinical use. Wikidata

Sarah E. MacPherson. Cognitive neuroscientist at the University of Edinburgh studying executive function and aging; linked Part B performance to processing speed, cortical thinning, and white-matter microstructure. ORCID - Faculty - Scholar

Ralph M. Reitan (1922–2014). Author of the seminal 1958 validity study that established the TMT as an indicator of organic brain damage and incorporated it into the Halstead–Reitan battery. Wikipedia - Wikidata

Timothy A. Salthouse. Originator of the processing-speed theory of cognitive aging (University of Virginia); his analysis quantified which cognitive abilities actually drive trail-making performance. Wikipedia - Faculty - Scholar - Wikidata

Tom N. Tombaugh (d. 2008). Neuropsychologist at Carleton University who produced the age- and education-stratified normative data most widely used to interpret TMT scores. Memorial

Frequently Asked Questions

What does the Trail Making Test measure?
Part A measures visual search and psychomotor speed; Part B adds cognitive flexibility, working memory, and set-shifting. The contrast between the parts isolates the executive switching demand from the shared perceptual-motor demand.

How long does the test take?
Administration is brief, usually two to five minutes total. A typical adult finishes Part A in about 30 seconds and Part B in about 75 seconds, though times lengthen considerably with age.

Why is Part B harder than Part A?
Part B requires alternating between a number sequence and a letter sequence, holding both in mind and switching between them, while Part A follows a single ascending series. The added switching and working-memory load makes B two to three times slower.

What are the B minus A and B over A scores for?
Both are derived indices that try to remove the visuomotor speed shared by the two parts, leaving a residual that reflects the executive switching cost. They guard against mistaking generalized slowing for an executive deficit.

Can I interpret a raw completion time by itself?
No. Performance changes greatly with age and education, so a raw time must be converted to a percentile or z-score using age- and education-stratified norms before it means anything clinically.

Is the Trail Making Test a test of the frontal lobes?
Part B is sensitive to fronto-executive function, but the underlying network is distributed across prefrontal, parietal, and subcortical white-matter regions. A poor score signals dysfunction without localizing it to a single site.

What conditions does the test help detect?
It is sensitive to traumatic brain injury, cerebrovascular disease, mild cognitive impairment, and dementia, all conditions that disturb processing speed and executive control. It flags problems for further assessment rather than diagnosing them.

How do digital versions differ from the paper test?
Tablet-and-stylus versions record the whole movement trajectory (pauses, hesitations before switches, pen lifts), so they decompose the single completion time into component sub-processes, potentially adding diagnostic detail the paper test cannot capture.

References