ConductScience VR

Virtual Hebb-Williams maze

People cross a virtual square field to a goal box, around walls rearranged for each new problem, as rodents do in the real maze.

ConductScience Editorial · Published November 15, 2026 · 6 references

STARTGOAL
Drawing · Hebb-Williams maze
Task
Find the way from a start box to a goal box at the opposite corner, around walls that change with each problem.1
First shown
Shore and colleagues first tested people on the mazes in 2001, in a computer-generated version.2
Across species
People and mice learned the mazes along strikingly similar curves.2
Fragile X
People with fragile X syndrome and Fmr1 knockout mice both showed spatial learning deficits on the mazes.3

What the virtual Hebb-Williams maze tests

Shore and colleagues built a computer-generated version of the Hebb-Williams mazes as a standard test, to compare people directly with mice run in the real mazes.2 They saw such a comparison as crucial if findings in genetically altered rodents are to be mapped onto human cognitive disorders.2

The rodent maze is a square field with a start box and a goal box at opposite corners.1 In the mouse maze, the field is 60 by 60 cm and its floor is marked into 36 squares.1 Barriers placed along the squares make each problem, and the squares also mark out error zones off the correct path.1 Moving the barriers gives problems of varying difficulty.1 The paradigm goes back to Hebb and Williams in 1946 and to Rabinovitch and Rosvold in 1951.1 The virtual field is a grid of 6 by 6 cells, like the physical maze.4

The first trial on a new layout is thought to reflect visual-spatial problem solving.1 Later trials reflect memory for the layout and the goal, which shows in how fast performance improves.1 Gabel and colleagues add that each maze calls on the visual cues of its walls and alleys, and on working and reference memory across trials.4

How a session runs

A session moves from practice mazes to test mazes, and each test maze runs for several trials in a row.4,1

  1. Practice moving and searching

    In their 2021 study of young children, Gabel and colleagues start with Maze 1, in which the goal, a red ball, is in view from the start box, as practice in moving through a maze.4 A training maze, Maze 5, then teaches the children to look around walls and in hallways to find the ball.4 Boutet and colleagues used an alley maze with the goal in view, then a T-maze that asks for a choice of path.1 People moved on after three trials in a row of under 30 seconds each.1

  2. Run each test maze several times

    Gabel and colleagues then run four test mazes, Mazes 6, 8, 11 and 12, each for six trials of up to 120 seconds.4 Each trial starts with the person facing a random direction in the start box, as in the physical maze used with animal models.4 Finding the red ball brings up a smiling cartoon star, and the person goes back to the start box for the next trial.4 If time runs out, yellow arrows appear on the floor and lead the person to the goal.4

  3. Score errors and time

    An error counts each time the person crosses into an error zone.1 The zones belong to the layout but are not visible to the person.4 Gabel and colleagues average the standardized error and time scores into one measure, performance efficiency, following Shore and colleagues.4

What the maze has shown in people

Fragile X syndrome

MacLeod and colleagues tested people with fragile X syndrome in the virtual mazes and Fmr1 knockout mice in the physical ones, on the same problems.3 Typically developing people of the same mental age improved across trials, with fewer errors and shorter times, while the fragile X group stayed at about the same level.3 The two groups did not differ on the first trial, so the deficit showed in learning across trials.1 MacLeod and colleagues suggest that affected people and mice show similar spatial learning deficits, which they attribute to the lack of the protein FMRP.3

Boutet and colleagues added a visual cue, a green star, to the virtual mazes for a new group of people with fragile X syndrome.1 The cue’s effects were inconsistent, with people finding the goal faster with it in Mazes 4, 5 and 9 and slower in Mazes 8 and 12.1

Reading impairment

Gabel and colleagues built on mouse work linking induced mutations in DCDC2, a candidate gene for dyslexia, to changes in visual-spatial learning.5 In 91 children aged 8 to 13, each given six trials of five virtual mazes, those with reading impairment did worse on the task.5 Of the 91, 18 were identified with a reading disorder.5 The authors suggest that deficits seen in genetic animal models of dyslexia carry over to people with reading impairment.5

A later study tested 42 children in kindergarten, before they could read, and again in second grade.4 Maze performance in kindergarten predicted later reading impairment.4 The link held with or without genetic risk, and was stronger with it.4 Struggling readers lacked a typical learning curve and were often inconsistent from one trial to the next.4 Typical readers, once they had found the goal, kept to the correct path on later trials.4 With no letters and no spoken answers in the task, the authors suggest it may suit remote testing and English language learners.4 The lead author, Gabel, has a US patent pending on the virtual maze as a tool for identifying dyslexia early.4

Blurred vision

Therrien and Collin used desktop renderings of the mazes to test how a range of blur levels affects route learning.6 Time to finish followed an S-shaped curve against blur, with its turning point near 2 diopters of defocus.6 The authors suggest that route learning by sight is fairly robust to blur, up to a threshold just above the limit for legal blindness.6

How it compares with the rodent maze

Because the rodent and virtual mazes pose the same problems, Boutet and colleagues describe people and mice as tested under equivalent conditions.3,1 A mouse runs the real field to a food cup in the goal box, while a person steers through the virtual field with a joystick or the arrow keys.1,4 For a mouse, time runs from the opening of the start box to its first bite of food.1 Boutet and colleagues compared the two species on errors, time and a third measure, inverse efficiency.1

Shore and colleagues found the learning curves of people and mice strikingly similar, and took this as support for the rodent model of human spatial memory.2 People learned faster than mice, and in both species females solved the problems less efficiently than males.2 In a first study of 37 people on four virtual mazes, Gabel and colleagues found performance consistent between people and mice.5 Boutet and colleagues report negligible species differences on Mazes 2, 4 and 5, with or without the visual cue.1 They suggest these three mazes could measure spatial learning in both species.1 For first-trial problem solving, Mazes 5 and 9 without cues gave the most consistent results.1

Results matched across species in some mazes and not others, and many conditions gave different results in people and mice.1 Boutet and colleagues gave mice five trials per maze and people three, to suit the limited attention span of the people with fragile X syndrome.1 Among MacLeod’s mice, Fmr1 knockouts made more errors than wild-type mice, though their times did not differ significantly.3 Boutet and colleagues note that some differences between mice and people may come from testing people in a virtual maze.1 They add that their virtual mazes, shown on a screen, test learning from sight alone, without the proprioceptive and motor input of moving through a real space.1

Build a virtual Hebb-Williams maze with ConductScience VR

We write the task software in house and build each Hebb-Williams maze to the protocol your lab plans to run.

  • The field and its grid, the start and goal boxes, the wall layout for each numbered problem you use and any visual cues
  • Practice mazes, trials per maze, time limits, guide arrows and the direction a person faces at the start
  • Each trial’s path, time to the goal and error zone entries, saved for analysis
  • A desktop screen with a keyboard or joystick, or a headset
  • The software alone, or a package with the headset and computer

We also build teaching versions for classes, where students test each other on the mazes.

Related pages

Questions

Is the virtual Hebb-Williams maze the same test as the rodent one?
It poses the same problems, in a field of 6 by 6 cells with start and goal boxes at opposite corners and walls moved for each problem.4,1 People see the maze on a screen and steer with a joystick or the arrow keys, while mice walk the real maze to a food cup.4,1
What does the virtual Hebb-Williams maze measure?
It records errors, each a crossing into an error zone off the correct path, and the time to reach the goal.1,4 The first trial on a layout is thought to reflect problem solving, and later trials memory for the layout.1
Does the task need reading or speech?
No, it uses no letters and asks for no spoken answers.4 Gabel and colleagues ran it with children in kindergarten, before they could read.4
Which groups has the virtual maze been tested with?
Studies have compared people with mice, and tested people with fragile X syndrome and children with and without reading impairment.2,3,5 Therrien and Collin also used it to study route learning under blur.6
Can the virtual Hebb-Williams maze run in a headset?
Yes, we build it for a headset or for a desktop screen with a keyboard or joystick. Boutet, Gabel and their colleagues showed the mazes full screen on a computer, steered with a joystick or the arrow keys.1,4

References

  1. Boutet I, Collin CA, MacLeod LS, et al. Utility of the Hebb-Williams Maze Paradigm for Translational Research in Fragile X Syndrome: A Direct Comparison of Mice and Humans. Frontiers in Molecular Neuroscience. 2018. doi:10.3389/fnmol.2018.00099
  2. Shore DI, Stanford L, MacInnes WJ, et al. Of mice and men: virtual Hebb-Williams mazes permit comparison of spatial learning across species. Cognitive, Affective & Behavioral Neuroscience. 2001. doi:10.3758/cabn.1.1.83
  3. MacLeod LS, Kogan CS, Collin CA, et al. A comparative study of the performance of individuals with fragile X syndrome and Fmr1 knockout mice on Hebb-Williams mazes. Genes, Brain and Behavior. 2010. doi:10.1111/j.1601-183X.2009.00534.x
  4. Gabel LA, Voss K, Johnson E, et al. Identifying Dyslexia: Link between Maze Learning and Dyslexia Susceptibility Gene, DCDC2, in Young Children. Developmental Neuroscience. 2021. doi:10.1159/000516667
  5. Gabel LA, Manglani M, Escalona N, et al. Translating dyslexia across species. Annals of Dyslexia. 2016. doi:10.1007/s11881-016-0125-3
  6. Therrien ME, Collin CA. Spatial vision meets spatial cognition: examining the effect of visual blur on human visually guided route learning. Perception. 2010. doi:10.1068/p5991

More virtual reality tests