The Aubert-Fleischl Paradox: Why Your Eyes Lie About the Speed of a Moving Target
When your eyes track a moving object, the brain underestimates its speed. New research reveals the neural mismatch between the moving retina and the motor signals that guide our gaze.

Elena Vasquez-Marsh · for The Unintuitive Universe · September 29, 2026
And it’s been measured. Every claim traced to the published research. Method & sources ↗
Hold your index finger out at arm's length, keep your eyes steady on a wall behind it, and sweep your finger slowly from left to right. Now, repeat the movement, but this time follow your fingernail closely with your gaze.
The finger appears to slow down the moment your eyes lock onto it.
Your finger did not decelerate. The change in pace is entirely an illusion generated by the visual cortex. First described by German psychophysiologist Hermann Rudolph Aubert in 1887 and Austrian physiologist Ernst von Fleischl-Marxow in 1882, the Aubert-Fleischl paradox exposes a fundamental math error in how the human brain calculates physical speed.
When you stare at a fixed point and an object sweeps across your field of view, its image slides over your retina. This physical sliding is known as retinal slip. The visual system reads this slip directly to calculate velocity. But when you track a target with your eyes—using what is called a smooth pursuit eye movement—the target's image is held nearly perfectly still on the fovea, the high-acuity center of your retina. Retinal slip drops to near zero.
To calculate how fast the target is actually moving under these conditions, the brain can no longer rely on retinal slip. Instead, it must reconstruct the speed by combining whatever tiny amount of retinal slip remains with a second signal: an estimate of how fast the eyeballs themselves are turning.
This is where the calculation fails. The brain chronically underestimates the speed of its own eye movements.
The Subtraction Problem
To build a stable map of the physical world, the motor system must communicate with the sensory system. When you decide to move your eyes, the brain’s motor regions generate an efferent command to the extraocular muscles. At the exact same time, a duplicate of this command—known as an efference copy or corollary discharge—is routed to the visual processing centers.
The visual cortex uses this efference copy to perform vector subtraction. In theory, if you subtract the estimated velocity of your eye movement from the motion detected on your retina, the remaining value should equal the true speed of the object in the physical world.
But the efference copy is not a perfect carbon copy. Neurophysiological studies show that the extra-retinal signal representing eye velocity is consistently scaled down. It behaves as if it is reporting a slower eye movement than what is actually occurring. When the brain subtracts this underestimated eye-speed signal from a near-zero retinal slip, the final calculated velocity of the target comes out lower than its true physical speed.
The result is a perceptual drop in speed. When an object is tracked, it is perceived as moving slower than when the eyes remain fixed and the object is allowed to sweep across a stationary retina.
Inside the Motion Centers
Functional neuroimaging and primate electrophysiology have traced the neural coordinates of this calculation to the superior temporal sulcus. Within this region, two distinct visual areas divide the computational labor: the middle temporal area (MT or V5) and the medial superior temporal area (MST).
Neurons in area MT are tuned primarily to retinal motion. They respond vigorously to the raw physical sweep of an image across the photoreceptor grid, but they do not account for what the eyes are doing. If an object is tracked perfectly so that its image remains stationary on the fovea, MT neurons show heavily attenuated activity, even though the object is moving rapidly through space.
By contrast, area MST receives direct extra-retinal signals, including the efference copy of the ocular motor command. MST neurons are responsible for translating retinal coordinates into head-centered coordinates, attempting to reconstruct true world-centered velocity.
Under the Bayesian models of sensory integration used to explain these illusions, the brain treats the motor system’s internal estimate of eye movement as a noisy, uncertain signal. Because the extra-retinal estimate is noisy, the brain biases its final speed calculation toward a "slow prior"—a hardwired assumption that things in the environment are generally stationary. When retinal slip is eliminated by smooth pursuit, this slow bias pulls the overall velocity estimate downward.
Real-World Limits of the Illusion
While the Aubert-Fleischl effect is highly robust in isolated laboratory conditions, its real-world impact is heavily shaped by the complexity of the surrounding environment.
Historically, experiments demonstrating the phenomenon utilized a single target moving across a completely black, uniform screen. In these empty-field environments, the lack of spatial reference points leaves the visual system entirely dependent on the flawed efference copy.
However, modern research utilizing three-dimensional virtual reality has demonstrated that the illusion can weaken or disappear entirely when a textured background is introduced. For instance, a study published in PLOS One by Björn Jörges and Laurence R. Harris (2025) measured speed perception during smooth pursuit against textured backgrounds. They found that when an object moves in front of a visible, structured environment, the relative motion cues between the target and the background provide the visual system with enough supplemental information to bypass the faulty motor estimate, preserving accurate speed perception.
When the visual field is rich with detail, the brain relies less on its own motor feedback and more on the geometric relationships within the scene itself. But whenever the background is dark or featureless, the brain is forced back onto its flawed internal bookkeeping, and the illusion of deceleration returns.
Measured.
This article is AI-generated (synthetic) content, produced by an automated editorial system with human direction and review. Every claim is traced to published, peer-reviewed sources.