What the test asks
When you change from one job to another, what slows you down: the change itself, or having two jobs in mind at all? A task-switching experiment separates the two, and a preparation interval shows how much of the cost you can prepare away. The second part measures something related with a pointer: how much longer a trail takes when you have to alternate between numbers and letters.
Part 1: task switching
A letter and a digit appear together in one cell of a 2 × 2 grid, for example G7 or 7G. In the top row you judge the letter (vowel = F or the left button, consonant = J or the right one), in the bottom row the digit (even = F or left, odd = J or right). The cell moves clockwise, so the task changes on every second trial and you always know which task is due.
- Single-task blocks. Only letters or only digits, 40 trials each, half at the start and half at the end, in mirrored order so that practice and fatigue fall on both about equally. The next item comes 675 ms after your answer, the midpoint of the two intervals below (our choice).
- Mixed blocks. 96 trials with 150 ms between your answer and the next item and 96 with 1,200 ms, in four blocks of 48 (short-long-long-short or the reverse).
- Each trial. The item stays until you answer, at most 3 s. After a mistake "Error" shows for 500 ms. Answers and letter-digit pairs are balanced across the cells, and one answer never comes more than three times in a row, a rule taken from Schmitter-Edgecombe and Langill.
- Practice. 12 letter, 12 digit and 16 mixed trials with feedback, required the first time.
Part 2: trails
Trails A: click or tap 25 circles in order, 1 to 25. Trails B: alternate, 1-A-2-B and so on up to 12-L-13, also 25 circles. The clock starts at your first tap on 1. A wrong circle flashes red three times and you carry on from the last correct one; a tap outside any circle is ignored. A practice of 8 circles comes before each trail the first time.
The layout is drawn afresh for each run, with no circles overlapping. Trails B is a mirror image of Trails A, so it looks different and its route is exactly as long. The route is also kept within a narrow range of lengths for the size of the field, and most steps do not go to the nearest circle, so that one run is about as hard as the next.
Things to try
- Compare the switch cost at 150 and 1,200 ms. How much goes away when you have time to prepare, and how much stays?
- Look at the mixing cost next to the switch cost. Are you slowed more by having two tasks in play or by the change itself?
- Use the 1,200 ms gap on purpose: look at the next cell and think of its task before the item appears.
- Do the trails once with a mouse and once by touch. The times will differ; does B/A?
- Repeat the trails over several days and watch the learning effect in History.
How the result is calculated
Only correct answers enter the times. The first trial of each block and every trial right after a mistake are left out, and so are answers under 100 ms. All values are differences of medians:
mixing cost = M(repeat in mixed blocks) − M(single-task blocks)
preparation effect = switch cost at 150 ms − switch cost at 1,200 ms
Here M is the median time in ms. A switch is the first trial after the task changed, a repeat the second trial on the same task. The error costs are the same differences for the share of wrong answers.
Each 95% interval is a percentile bootstrap with 2000 repeats: the page draws new sets of times from yours, with replacement, and keeps the middle 95% of the differences. In simulations (1 November 2026) with a true switch cost of 150 ms and a mixing cost of 100 ms, the intervals held the true value 95 to 98% of the time. Their width depends on how steady your times are. With a spread (SD) of about 134 ms and 3% attention slips they were typically 87 ms wide for the switch cost, 91 ms for the mixing cost and 182 ms for the preparation effect; with a spread of about 42 ms and no slips, 29, 31 and 60 ms. So go by the interval of your own run: a difference between two runs that is smaller than it can be chance.
For the trails, the time runs from the first tap on 1 to the tap on the last circle, and the page also shows the time of each step. The ratio is the time for B divided by the time for A. One run gives no interval for it.
Compared with published values
These are landmarks from other groups and tasks, so no rank is computed from them.
| Study | Value |
|---|---|
| Wasylyshyn, Verhaeghen and Sliwinski 2011: 26 articles, 36 groups, many switching tasks | A mixed block adds about 141 ms in younger adults. A switch is about 43% slower than a repeat in the same block. |
| Rogers and Monsell 1995, as described by Schmitter-Edgecombe and Langill 2006 | From 150 to 600 ms of preparation the switch cost fell from over 200 ms to about half. Beyond 600 ms it fell no further. |
| Woods et al. 2015, Experiment 1: 165 people aged 18 to 82, digital trails with a mouse, 60 Hz, equal routes | A 37.4 s (SD 15.5), B 60.8 s (SD 29.5), B − A 23.4 s (SD 25.3), B/A 1.69 |
In the data of Woods and colleagues the ratio B/A was not affected by age, education, sex or computer use, while the times were affected by age and computer use, which is why this page leads with the ratio. Their participants moved the mouse from circle to circle and clicked inside each one, with the path shown on the screen; here you click or tap the circles and no path is shown while you move, so the times are not directly comparable. In the meta-analysis the mixing cost grew with age beyond general slowing, and the switch cost did not.
Why there are no paper norms and no percentile
Paper trail norms disagree widely. Across the 13 large studies in Table 1 of Woods et al., average times run from 23.4 to 70.2 s for A and from 54.3 to 157.7 s for B, partly because of age and partly because of how the test was given and timed. In a common printed version the route of B is also 32% longer than that of A. A click on a screen cannot be read against those numbers, so they are not shown as a standard.
Repeatability is the other reason. In Woods et al. Experiment 2, 50 people came back for three weekly sessions: the intraclass correlation of the total time, as z-scores adjusted for age and computer use, was 0.87 for A and 0.85 for B but only 0.56 for B minus A, and times fell by 12% on A and 20% on B, mostly from the first session to the second. Switch and mixing costs are differences too, so read them with their intervals and over several runs.
Where the tasks come from
The letter and digit task follows the predictable design of Rogers and Monsell (1995), as described by Schmitter-Edgecombe and Langill (2006), whose letter and digit sets it uses. In that reading part of the cost is a reconfiguration of the task set that you can do in advance, which is why preparation helps, and a residual part stays even after long intervals. The meta-analysis calls the two costs local (switch) and global (mixing): the global cost as keeping and scheduling two task sets, the local cost as changing from one to the other.
The Trail Making Test was first published by the US Army in 1944 and is in the public domain. The digital version with equal routes is the work of Woods and colleagues. The layouts on this page are its own, not those of any published form.
Limits
- Mouse, touch and paper differ. History keeps mouse and touch apart, and no result here can stand in for a paper test.
- Your device. The screen, keyboard and browser add a delay to every time. A constant delay cancels out in a difference of medians, but not in a trail time. Below 50 Hz or with uneven frames, which remote desktop and power-saving modes often cause, the page warns you first and offers Run anyway: the run is marked less precise and kept apart in History.
- Keyboard. The trails need a pointer, because the task is to find and point at a circle. With only a keyboard you can run the first part.
- Small screens. The circles are at least 44 px across from a 360 px wide screen; on a narrower one they get smaller and only their touch area stays 44 px.
- Studying the layout. The clock starts at your first tap, so looking long at the circles beforehand would shorten the time. Start as soon as you see them.
- A hidden tab. The trial in progress is discarded and repeated, and a trail in progress starts again. The run waits for Resume.
- Not a medical test. This is not a clinical trail-making assessment and gives no cut-off for brain injury or dementia.
Controls
- Keyboard: F and J answer in Part 1. Enter or Space on the focused button goes on; Esc stops a run. The trails cannot be done with a keyboard.
- Touch and mouse: the two buttons under the field answer in Part 1, the left one as F and the right one as J. In Part 2 click or tap the circles.
- Pauses: for 0.7 s after the end of a block or part, and after Stop, the buttons are dimmed and ignore keys still pressed from the task. A second click within 0.7 s of a start does not stop the run, so a double click is safe.
- Language: the link at the end of History switches the app to Slovak.
Sources
- Wasylyshyn, C., Verhaeghen, P., & Sliwinski, M. J. (2011). Aging and task switching: A meta-analysis. Psychology and Aging, 26(1), 15–20. Full text (PMC4374429)
- Woods, D. L., Wyma, J. M., Herron, T. J., & Yund, E. W. (2015). The effects of aging, malingering, and traumatic brain injury on computerized trail-making test performance. PLoS ONE, 10(6), e0124345. Full text (PMC4465490)
- Schmitter-Edgecombe, M., & Langill, M. (2006). Costs of a predictable switch between simple cognitive tasks following severe closed-head injury. Neuropsychology, 20(6), 675–684. Full text (PMC1779821)