How the Trail Making Test Measures Cognitive Function

The Trail Making Test is one of the most widely used neuropsychological tests in the world, given in clinical settings to measure processing speed, visual attention, and mental flexibility. It comes in two parts: Part A asks the person to connect numbered circles in order (1-2-3-4…), while Part B raises the difficulty by requiring alternation between numbers and letters (1-A-2-B-3-C…). The difference in performance between these two halves is where much of the test’s diagnostic power lies, and the test shows up in evaluations for everything from dementia to traumatic brain injury to fitness to drive.

How the Test Works

The traditional version uses paper and pencil. A sheet of paper has 25 circles scattered across it, each containing a number (Part A) or a number or letter (Part B). You draw a line connecting the circles in the correct sequence as quickly as you can, without lifting the pencil. If you make a mistake, the examiner immediately points it out and you correct it before continuing. The clock keeps running through errors, so mistakes cost you time even though they get corrected on the spot. The primary score is simply how many seconds it takes to finish each part.

Part A is fairly straightforward. You scan the page, find the next number, and draw a line to it. It taxes basic visual scanning and motor speed. Part B is where things get interesting: you have to hold two sequences in mind simultaneously and switch between them. That added demand is why Part B takes longer for everyone, but it takes disproportionately longer for people with certain kinds of cognitive impairment.

What Part B Actually Measures

There has been genuine debate about what cognitive ability Part B really taps into. The intuitive answer is “cognitive flexibility,” the capacity to shift between two mental sets. Research supports this: one study found that Part B performance was more closely tied to cognitive flexibility than to the ability to simply maintain a rule once you’ve started following it.¹ But the picture is muddied by the fact that Part B also differs from Part A in motor demands and visual complexity, not just in the mental switching required.²

A large structural analysis found that trail-making performance largely reflects individual differences in processing speed and fluid cognitive ability, with the balance between those two depending on which scoring method you use.³ So when a clinician says “the TMT measures executive function,” that’s a useful shorthand, but the test is really capturing a blend of speed, scanning ability, sequencing, and set-switching all at once.

What Happens in the Brain During the Test

Brain imaging studies have tracked what fires up when someone performs the TMT inside a scanner. The general finding is that both parts activate widespread networks across the frontal and parietal lobes, along with visual cortex and motor regions, which makes sense for a task that combines looking, planning, and drawing.⁴

When researchers compare brain activity during Part B versus Part A, the extra activation is mainly left-sided. An fMRI study found distinct left dorsolateral and medial frontal activity during Part B, consistent with the idea that left frontal regions handle the set-switching demands. But the study also found activity in the left temporal gyrus, reinforcing that the brain regions involved in the TMT are not restricted to the frontal lobe alone.⁵ A study of older adults found a similar pattern: Part B compared with Part A produced increased activation in the left inferior parietal lobe, left frontal gyri, the left insular cortex, and the middle cingulate gyrus.⁶ The left-hemisphere dominance probably reflects the verbal labeling involved in tracking letters alongside numbers.

Scoring Beyond Raw Time

The simplest scores are Part A time and Part B time. But clinicians sometimes calculate derived scores to try to isolate executive function from pure speed. The two most common are B minus A (subtracting Part A time from Part B time) and B divided by A (a ratio). The logic is that subtracting or dividing out the “speed” component of Part A should leave a purer measure of the switching component. These derived scores do enhance measurement of executive functioning in some contexts, but the subtraction score (B minus A) can produce misleading estimates of executive dysfunction in clinical samples.⁷

A study of people with traumatic brain injury found that derived indices did not add unique interpretive value beyond what the raw Part A and Part B times already provide.⁸ This remains a point of discussion among neuropsychologists: the derived scores are elegant in theory, but in practice, their added benefit is inconsistent. Most clinicians still rely heavily on the raw completion times interpreted against normative tables.

How Age and Education Affect Scores

Performance on both parts slows with age and is worse at lower levels of education. This is not a subtle effect. A large normative study stratified norms across 11 age groups and two education levels because the differences were substantial enough to matter for clinical interpretation.⁹ A 70-year-old with a high school education taking 90 seconds on Part B might be performing normally, while a 40-year-old college graduate with the same time would raise red flags. Without age- and education-corrected norms, you can easily misinterpret a normal older adult’s score as impaired, or miss genuine impairment in a younger, well-educated person.

Screening for Dementia and Mild Cognitive Impairment

The TMT is commonly included in dementia evaluations. Research shows that both completion time and error rate on Part B differ across people who are cognitively healthy, people with mild cognitive impairment (MCI), and people with dementia, and that time and errors each carry independent diagnostic information.¹⁰ Errors are sometimes overlooked in clinical practice because the standard scoring focuses on time, but a person who completes Part B quickly while making several uncorrected errors may be showing a different pattern of impairment than someone who is simply slow.

A study using the Chinese version of the TMT found strong diagnostic accuracy for separating Alzheimer’s disease from healthy controls, with Part B showing around 83% sensitivity and 92% specificity at a particular time cutoff.¹¹ The test was less sensitive for distinguishing MCI from healthy aging, which is consistent with MCI being a subtler condition that sits closer to normal cognition on most measures. Imaging research has linked poor Part B performance in people with MCI and early Alzheimer’s to dysfunction in the anterior cingulate, striatum, and thalamus, brain structures involved in attention and cognitive control.¹²

Traumatic Brain Injury

People recovering from traumatic brain injury (TBI) tend to perform worse on both parts of the TMT compared with healthy controls, but Part B is more consistently affected.¹³ One study found that among TBI patients, Part A times were mostly within the normal range, while Part B showed small but significant increases, suggesting that the switching demands of Part B expose cognitive deficits that simple visual tracking does not. Movement velocities were also slightly slower in TBI patients on both parts, hinting at a motor component to the slowdown beyond just the cognitive demands.

When TBI patients’ scores were compared with those of people with schizophrenia, TBI patients actually performed worse on raw Part A and Part B times. However, the ratio and difference scores were similar between the two groups, suggesting a comparable underlying executive deficit despite different conditions.¹⁴ Research on bilingual TBI survivors also found group differences on both TMT-A and TMT-B compared with healthy controls, particularly at six months post-injury.¹⁵

Psychiatric Conditions

The TMT sees heavy use in psychiatric assessment, especially in schizophrenia research. A meta-analysis confirmed that people with schizophrenia perform significantly worse than healthy controls on both Part A and Part B, reflecting impaired processing speed and cognitive flexibility.¹⁶

Where it gets clinically useful is in distinguishing patterns. A study comparing schizophrenia, major depression, and healthy controls found that both patient groups were slower than controls, but only the schizophrenia group made significantly more errors, particularly tracking errors, suggesting a greater degree of cognitive disorganization beyond simple slowness.¹⁷ Another study found that depressed patients showed impairment mainly on Part B, not Part A, while schizophrenia patients were slower on both.¹⁸ These patterns can help clinicians understand the nature of cognitive difficulties in different psychiatric presentations, even if the TMT alone is not enough to make a diagnosis.

Predicting Real-World Functioning

A fair question about any neuropsychological test is: does performance on it predict anything that matters in everyday life? For the TMT, the answer is yes, particularly for Part B. In community-dwelling older adults, Part B performance made a significant and unique contribution to predicting how well someone could manage instrumental activities of daily living, things like managing finances, preparing meals, and handling medications.¹⁹

In Parkinson’s disease, the sequencing element of the TMT turned out to be the strongest predictor of daily functioning, more so than scanning or cognitive flexibility.²⁰ This is a meaningful finding because it tells clinicians and families that when a person with Parkinson’s starts struggling with the TMT, it may signal emerging difficulty with managing daily tasks independently.

Driving Fitness

One of the more practical applications of the TMT is in assessing whether older adults or cognitively impaired patients can still drive safely. Research has explored this, but the results are mixed enough to warrant caution. One study found that TMT-A scores in a range of 68 to 90 seconds identified a gray zone where additional on-road testing was warranted, and that inability to complete Part A in under 48 seconds or Part B in under 108 seconds helped distinguish unsafe from safe or marginal drivers among cognitively impaired individuals.²¹ But a high percentage of cognitively impaired participants could not even complete Part B within the allowed time, which limited its usefulness for that group.

Another study sounded a clearer warning: while older patients with a Part A time of 54 seconds or more, or a Part B time of 150 seconds or more, had roughly three times the risk of poor on-road performance, the test’s specificity was too low to justify driving restrictions on its own. In a screening setting, it would have flagged nine drivers out of ten unnecessarily.²² The takeaway for families and clinicians is that a poor TMT score is a reason to pursue further evaluation of driving ability, not a reason to take away the keys.

Reliability and Practice Effects

For a test that gets used as frequently as the TMT, reliability matters. If you give someone the test twice a week apart, do they get a similar score? Studies of alternate forms (redesigned layouts with the same task structure) show strong retest reliability, with coefficients ranging from 0.76 to 0.89 for Part A and 0.86 to 0.94 for Part B.²³ Part B tends to be more reliable than Part A, likely because it takes longer and involves more varied cognitive demands, which reduces the influence of random variability.

Practice effects are a concern when the test is given repeatedly, such as in longitudinal monitoring of cognitive decline. Research in children found significant improvement in raw TMT-A and TMT-B scores across a one-year interval, though the derived ratio and difference scores stayed stable across test sessions.²⁴ The stability of derived scores across retesting is one argument in their favor, even though their added diagnostic value over raw scores is debated.

Digital Versions and Their Tricky Equivalence

Tablets and computers offer obvious advantages for administering the TMT: automated timing, easier data storage, and the potential for remote testing. But moving from paper to screen is not as simple as recreating the same circles on a different surface. One study comparing an iPad version to the paper original found that the tablet Part A had poor test-retest reliability, while Part B was more reliable on the tablet. Conversely, tablet Part A showed reasonable agreement with the paper version’s scores, but Part B did not. The authors cautioned that electronic versions may end up measuring somewhat different constructs than the paper test.²⁵

Other studies have been more optimistic. A validation of a tablet-based version found strong correlations between digital and paper scores, though participants were generally faster on the tablet. The authors concluded that with new normative data, the TMT can be successfully adapted to digital platforms.²⁶ A separate study of a computerized TMT found high reproducibility and strong correlations with the paper version, with favorable sensitivity and reliability.²⁷ The emerging consensus is that digital TMTs can work well, but you cannot simply apply paper-based norms to digital scores. Any clinic switching to a tablet version needs to use norms developed specifically for that platform.

Detecting Faked or Low-Effort Performance

In forensic and compensation settings, there is always the question of whether someone is giving their genuine best effort. The TMT has been studied as an embedded performance validity measure, meaning clinicians can look at the scores themselves for patterns suggesting poor effort, without needing a separate validity test. Among veterans, TMT-derived validity indicators showed moderate ability to flag invalid performance, with sensitivity ranging from about 32% to 48% at specificity levels above 91%.²⁸

A study of brain injury rehabilitation patients found that Part A completion time had some value as a validity indicator, but Part B time showed poor sensitivity and positive predictive value for detecting low effort. The authors were clear that no TMT-based validity indicator should replace dedicated effort tests during neuropsychological assessment.²⁹ The TMT can raise a flag, but it cannot definitively distinguish someone sandbagging from someone who is genuinely impaired.

Cultural Adaptations

The standard TMT uses the Latin alphabet, which creates an obvious problem for populations that use different writing systems. Part A (numbers only) transfers reasonably well, but Part B (numbers and letters) essentially becomes a different test if the person is not fluent in the Latin alphabet. Researchers have developed alternatives like the Color Trails Test, which uses colored circles instead of letters, and the Shape Trail Test, developed specifically for Chinese-speaking populations.³⁰ These variants try to preserve the core cognitive demand of Part B, alternating between two sequences, while removing the dependence on a particular alphabet. This is an active area of development, and clinicians working with multilingual or non-English-speaking populations need to be aware that the standard TMT may not produce valid Part B scores in those groups.

Alcohol and the TMT

The TMT’s sensitivity to acute cognitive changes makes it useful for studying substance effects. A field study of underage drinkers found that breath alcohol levels predicted performance on Part B but not Part A, consistent with the idea that alcohol impairs the higher-order switching demands of Part B while leaving basic motor speed and scanning relatively intact. Perhaps more striking, years of drinking experience independently predicted worse performance on both parts, above and beyond the effects of current intoxication.³¹ That finding hints at cumulative cognitive costs of regular drinking that show up even when the person is sober during testing.

Use in Children and ADHD

Although the TMT was originally developed for adults, it is used in pediatric neuropsychology as well. In children and adolescents with ADHD, the Part B minus Part A difference, which isolates the switching component from basic speed, tends to be larger than in typically developing peers, and it correlates with parent-reported symptom severity. This switching-cost difference decreases with age, suggesting that the executive control demands captured by Part B may be especially challenging for younger children with ADHD and partly normalize as they mature.³² Pediatric norms are essential here because children’s TMT times change rapidly with development, and using adult norms would produce meaningless results.