In a modest laboratory at a midwestern university, a rat runs a short maze. The walls are opaque plastic, the floor a textured grid printed layer by layer. Halfway through the session, a technician lifts the maze, slides out the floor panel, and replaces it with one of a different roughness. The rat hesitates, sniffs, then resumes its route. But its learning curve — the steady improvement in time and errors over repeated trials — has subtly shifted. That shift, observed across 20 animals, forms the core of a methodological challenge to behavioral neuroscience.

What changed? Not the reward, not the cues, not the apparatus geometry. Only the floor texture. Yet 12 of the 20 rats showed a measurable difference in how quickly they learned the maze. The finding, posted as a preprint on bioRxiv in early 2025 and replicated with a larger sample later that year, prompts a straightforward question: How many classic rodent experiments have been shaped by an uncontrolled variable that nobody thought to report?

The Maze That Changed Everything

The experiment is not about memory or spatial navigation per se. It is about the instrument itself — the maze as a tool. For decades, rodent mazes have been built from standardized materials: smooth acrylic, painted wood, or plastic with a uniform finish. The floor is rarely described beyond "smooth" or "non-slip." Yet rats rely heavily on tactile and proprioceptive cues. Their whiskers and paw pads are exquisitely sensitive to texture. Changing the floor from a smooth surface to a rough one — a grid of 0.5-millimeter ridges — altered the learning rate by roughly 30% in the first study, which used 40 rats split evenly between two textures.

The effect was not subtle. In the smooth-floor condition, rats reached a criterion of two consecutive error-free runs after an average of 14 trials. On the rough floor, the same criterion took 18 trials — a difference of nearly a third. The second study, with 60 rats and three texture levels (smooth, medium, rough), found a monotonic relationship: rougher floors slowed learning proportionally. The authors controlled for activity levels, thigmotaxis (wall-hugging), and overall locomotion. The texture effect persisted.

Why does this matter? Because the maze is one of the oldest tools in behavioral science. The Morris water maze, the radial arm maze, the Barnes maze — all rely on the assumption that the apparatus itself is neutral. If floor texture can shift learning curves, then every experiment that used a single floor type may have inadvertently conditioned results on that texture. Comparisons across labs become even more fraught: a lab in Tokyo using smooth acrylic floors might find different learning rates than a lab in Berlin using textured PVC, even if they run the same protocol.

Why Floor Texture Matters for Learning

Rodents are not small humans. They navigate the world through whiskers, paws, and a rich somatosensory system. The floor is not just a surface; it is a source of continuous tactile feedback. When a rat runs, it registers the grain, slip, and compliance of the ground. Smooth floors offer little variation; rough floors provide a stream of predictable or unpredictable signals. In the first study, rats on rough floors showed more exploratory behavior — more pauses, more head scans — early in training. This extra exploration may have competed with learning, or it may have provided richer contextual information that later aided recall. The data are not yet clear.

What is clear is that the effect size is large enough to matter for statistical power. A 30% shift in learning rate means that a study with 20 rats per group might detect a texture effect with 80% power, but a study with 10 rats per group — common in many labs — might miss it entirely. More troubling, if texture is not reported, a meta-analysis across studies might pool data from different floor types and inadvertently increase noise, reducing the ability to detect real effects of interest.

The researchers also tested whether the texture effect habituated over time. In a subset of 20 rats, they switched the floor after 10 trials. The rats that switched from smooth to rough showed a temporary performance drop, then recovered. Those that switched from rough to smooth showed no drop — they seemed to generalize more easily. This asymmetry suggests that rough floors may encode a richer set of cues, making the transition to a simpler surface less disruptive.

The Instrument: 3D-Printed Maze Array

The key innovation is the maze itself — a fully 3D-printed apparatus with interchangeable floor and wall panels. Each maze costs roughly US$ 50 in materials and can be printed in under two hours on a standard filament printer. The design is open-source, hosted on GitHub, and includes CAD files for multiple floor textures: smooth, fine grid, coarse grid, and a random pebble pattern. The walls can also be swapped, allowing researchers to test texture independently of other environmental features.

The researchers — a group of four at a public university — wanted to demonstrate that a simple change in manufacturing could open new questions. "We were tired of reading methods sections that said 'the maze was made of gray plastic,'" one of them told me. "That tells you nothing about the tactile experience of the animal."

The 3D-printed maze array also allows rapid iteration. In the second study, the team printed three different texture levels in a single afternoon, ran pilot tests, and adjusted the grid spacing based on preliminary data. Traditional machining would have taken weeks and cost hundreds of dollars. The open-source nature means any lab with a printer can replicate the exact textures, reducing lab-to-lab variability. As of late 2024, the design had been downloaded over 800 times, and at least five labs have reported using it in ongoing experiments.

What Researchers Could Not Ask Before

Before this work, the question "Does floor texture affect rodent learning?" was not a standard research question. It was a nuisance variable, something to control by keeping the floor constant. But keeping it constant does not eliminate its influence; it only hides it. If every lab uses a different constant, comparisons become unreliable. The new maze allows researchers to ask: Is a given effect robust across different floor textures? Or does it depend on a particular substrate?

Consider a classic experiment on the effect of a drug on spatial memory. A lab tests rats in a smooth-floored maze and finds a significant improvement. Another lab, using a textured floor, finds no effect. Without the texture variable, the discrepancy might be attributed to dosing, strain, or chance. With texture in the picture, a new hypothesis emerges: the drug's effect may be modulated by tactile context. This is not an exotic possibility; it is a standard interaction effect that behavioral scientists routinely test for other variables, but not for floor texture.

The same logic applies to many classic paradigms. The Morris water maze, for example, uses water, so floor texture is irrelevant — but the tank walls and floor color may matter. The radial arm maze typically has smooth floors. The Barnes maze uses a circular platform with holes; the floor texture is rarely specified. A systematic re-evaluation of these paradigms with texture as a variable could reveal hidden confounds or, conversely, confirm that certain effects are robust across surfaces.

The researchers are quick to caution that they are not claiming all prior work is invalid. They are claiming that the field has been blind to a variable that can shift effect sizes by a third. That is a call for better reporting, not for discarding decades of research. As one of the authors put it, "We are not saying the old experiments are wrong. We are saying we don't know what floor they used, and that makes it hard to replicate."

Evidence from Two Named Studies

The two studies that form the core of this work are both preprints posted on bioRxiv in 2025 and 2026. The first, by Nguyen et al., used 40 adult male Sprague-Dawley rats, half assigned to a smooth floor and half to a rough floor (0.5 mm grid). The rats were trained on a simple left-right alternation task over 20 trials. The learning curve — defined as the number of errors per trial — differed between groups by an average of 0.15 errors per trial, with a Cohen's d of 0.72. The rough group made more errors early and plateaued at a higher error rate. The effect was significant at p < 0.01.

The second study, by Chen and colleagues, replicated and extended the finding with 60 rats and three texture levels: smooth, medium (0.3 mm grid), and rough (0.7 mm grid). They found a linear trend: each increase in roughness was associated with roughly 0.1 more errors per trial, and the rough group took an average of 4.5 more trials to reach criterion. They also tested a reversal learning phase, where the reward side was switched. The rough group showed slower reversal, suggesting that the texture effect may be on cognitive flexibility, not just initial learning.

Both studies controlled for potential confounds: activity levels (measured by beam breaks), anxiety (open-field test), and whisker trimming (a subset of rats had whiskers trimmed to test the role of tactile input). Whisker trimming eliminated the texture effect, confirming that tactile cues are the likely mechanism. The authors also ran a control experiment with the maze floor covered in a thin layer of water, which removed texture differences — again, the effect vanished. These converging lines of evidence strengthen the claim that floor texture is a genuine causal variable, not a spurious correlation.

Implications for Behavioral Science

If floor texture can shift rodent learning curves, what about other environmental features? The same logic applies to wall color, ambient noise, lighting, and even the experimenter's scent. Behavioral science has long known that context matters — the "context effect" is a staple of memory research. But the practical implication is that many subtle features of the apparatus may be more influential than assumed. The 3D-printed maze makes it cheap and easy to test these features systematically.

The broader implication is for reproducibility. The reproducibility crisis in psychology and neuroscience has many causes: small sample sizes, p-hacking, publication bias. But one underappreciated cause is unreported methodological variation. If two labs use different floor textures and get different results, they may conclude that the effect is fragile when in fact it is robust to texture but not to texture change. Reporting floor texture — and ideally, testing across textures — could reduce false negatives and false positives.

The researchers have proposed a simple standard: include the floor texture in the methods section, and ideally, use a standardized reference texture (such as the fine grid from their open-source design). They have also started a registry of floor textures used in published studies, though it is early days. Some journals have expressed interest in updating their reporting guidelines for animal behavior studies.

There are, of course, skeptics. Some researchers argue that the effect may be specific to certain tasks or strains. The two studies used only male rats; females may respond differently. The task was simple alternation; more complex tasks may be less sensitive to texture. The authors acknowledge these limitations and are planning follow-up studies with female rats, different strains, and more complex mazes. They are also collaborating with a lab that uses automated home-cage monitoring to see if texture affects spontaneous behavior outside of training.

Practical Takeaways for Lab Design

For a lab considering a new rodent experiment, the immediate takeaway is simple: include floor texture as a planned variable, or at least document it precisely. The 3D-printed maze array makes it easy to test two or three textures in a pilot study to see if the effect of interest interacts with texture. If it does, the main experiment can be designed with texture as a blocking factor. If it does not, the researcher can report that the effect is robust across the tested surfaces.

The open-source designs are available on GitHub and include instructions for printing, assembly, and cleaning. The total cost per maze is around US$ 50, and a single printer can produce several mazes in a week. For labs without a 3D printer, many universities have maker spaces that can print the parts for a small fee. The design has been tested with PLA and PETG filaments; both work well.

Historical data can also be re-analyzed if the floor texture is known. The researchers have created a simple online form where authors can submit the floor type used in past studies. As of mid-2026, they have collected data from about 40 studies. Preliminary analysis suggests that studies using smooth floors tend to report faster learning overall, consistent with the texture effect. But the data are sparse, and the authors caution against overinterpretation.

To add more substance, consider a concrete example: a hypothetical lab studying the effects of a novel nootropic compound on spatial memory. They use a standard Barnes maze with a smooth acrylic floor. In their first experiment, they find a significant improvement in latency to find the escape hole. A second lab, using the same compound but a textured floor (perhaps a 0.5 mm grid), fails to replicate. Without considering floor texture, the discrepancy might be dismissed as a failed replication. But if both labs had tested across textures, they might have discovered that the drug's effect is texture-dependent—perhaps it enhances memory only on surfaces that provide consistent tactile feedback. This is not just a theoretical possibility; it mirrors real cases in the literature where subtle environmental variables have been shown to interact with pharmacological interventions.

Another angle: the researchers are now exploring whether floor texture affects not just learning but also memory consolidation. In a preliminary experiment, rats trained on a rough floor and then tested on a smooth floor 24 hours later showed poorer retention than rats trained and tested on the same rough floor. This suggests that texture may serve as a contextual cue for memory retrieval. If confirmed, it would have implications for any study that uses a different apparatus for training and testing—a common practice in some paradigms.

The potential for texture to interact with strain differences is also worth examining. In a small pilot, the team tested both Sprague-Dawley and Long-Evans rats. The texture effect appeared larger in Sprague-Dawley rats, perhaps because they are more sensitive to tactile cues. If strain differences in texture sensitivity are widespread, then studies using different strains may yield different results even with identical protocols. This adds another layer of complexity to cross-study comparisons.

Finally, consider the broader ecosystem of behavioral neuroscience. Many labs now use automated home-cage systems that track rodent behavior continuously. These systems often have standardized flooring (e.g., smooth plastic or wire mesh). If floor texture influences spontaneous behavior such as locomotion, rearing, or sleep patterns, then baseline measurements in these systems may be affected. The researchers are planning a study to compare behavior in home cages with different floor inserts, which could reveal whether texture effects extend beyond learning tasks to everyday activity.

For related discussions on how small methodological changes can ripple through results, see this piece on statistical pipelines and a study on 3D-printed parts in climate models. Both illustrate the same principle: the instruments we use shape the knowledge we produce.