The Reafference Principle

The claim

The nervous system predicts the sensory consequences of its own movement

In 1950 Erich von Holst and Horst Mittelstaedt published Das Reafferenzprinzip in Naturwissenschaften, volume 37, issue 20, pages 464 to 476 (von Holst and Mittelstaedt 1950). The claim is short enough to state in one sentence and large enough to reorganize a century of physiology. Every time the central nervous system issues a motor command, it retains a copy of that command. The copy is the efference copy, Efferenzkopie in the original. Its job is to predict the sensory feedback the movement will produce. When prediction and feedback match, the system stays quiet. When they fail to match, the difference is the signal.

Ask what that commits you to. If the nervous system holds a forecast of its own movement, then sensation is never raw. Every incoming signal arrives somewhere that already has an expectation waiting for it, and what reaches awareness is the remainder. Perception becomes subtraction. The body is not reading the world so much as checking the world against a prediction it wrote a few milliseconds earlier. Von Holst and Mittelstaedt never used the phrase predictive brain. They built the machine anyway, out of flies, fish, and a diagram with one arrow pointing the wrong way. The paper ran to thirteen pages, and it changed what a feedback loop was allowed to mean.

Riga to Seewiesen

Erich von Holst spent thirty years arguing that the nervous system is not a reflex machine

Erich von Holst was born in Riga on 28 November 1908 and died at Herrsching am Ammersee on 26 May 1962, aged 53. He studied at Kiel, Vienna, and Berlin, and took his doctorate in 1932 under Richard Hesse with a dissertation on the central nervous system of the earthworm. He held a fellowship at Albrecht Bethe's institute in Frankfurt from 1933, worked at the Naples Zoological Station from 1934 to 1936, and habilitated at Gottingen in 1938. In 1946 he became professor of zoology and director of the Zoological Institute at Heidelberg.

The institutional line matters because of where it ends. In 1949 he helped found the Max Planck Institute for Marine Biology at Wilhelmshaven. In 1950 he set up a research station for comparative ethology at Buldern castle. In 1954 the two operations merged into the Max Planck Institute for Behavioral Physiology, and from 1955 von Holst directed it at Seewiesen near Starnberg, with Konrad Lorenz as his deputy. He was elected to the Leopoldina in 1957 and to the academies at Gottingen, Heidelberg, and Munich. In his own hours he built rubber powered flying models of birds and pterosaurs and designed asymmetric violas. The man who wrote the reafference principle spent his evenings on flight and resonance. He was 53 when he died. The institute he built outlived him by 37 years and closed in 1999.

The eel

Cut the sensory nerves and the rhythm keeps running

Before the efference copy there was the eel. Through the 1930s von Holst worked on the locomotion of fishes, and his decisive preparation removed sensory input from the spinal cord. A deafferented eel still swam. The undulating wave still traveled down the body in order and at rate, with no sensory signal available to trigger the next segment. Niko Tinbergen cited these investigations in The Study of Instinct (Tinbergen 1951). Von Holst had summarized the fish work in 1937 (von Holst 1937), and Konrad Lorenz returned to the deafferented spinal eel for years afterward whenever he needed a clean example of behavior that owes nothing to stimulation. The result cut directly against the reigning model, the reflex chain, in which every movement is set off by the sensation the previous movement produced.

Ask what that commits you to. If rhythm survives the loss of feedback, then the pattern is generated inside, and sensation is not the cause of movement but the correction of it. The nervous system is an oscillator first and a responder second. That single reversal is the difference between a body that waits to be pushed and a body that keeps its own time. Von Holst had already shown that the central nervous system runs, and runs on its own, before he asked the harder question of what it does with the sensations its running produces.

Magnet effect

Two rhythms in one body pull on each other

In 1939 von Holst published Entwurf eines Systems der lokomotorischen Periodenbildungen bei Fischen, an outline of a system of locomotor rhythm formation in fishes (von Holst 1939). Watching fins beat at different rates on the same animal, he named two forces. The maintenance tendency, Beharrungstendenz, is the drive of each oscillator to hold its own frequency. The magnet effect, Magneteffekt, is the pull one oscillator exerts on another, dragging it toward a shared rhythm. Where the two balance you get what he called relative coordination: rhythms that are neither locked together nor fully independent.

Those are not metaphors borrowed from physics. They are coupled oscillator behaviors, described in a fish tank a generation before they became the formal core of movement science, and they are still cited in current work on interlimb coordination and entrainment. Relative coordination is what a living system looks like from the inside: partial agreement, drifting phase, constant negotiation. Hold it alongside the reafference principle and the whole program comes into focus. The body generates its own rhythms, the rhythms pull on one another, and the sensory world is judged against what those rhythms predicted. Frequency, phase, coupling strength. The vocabulary was in place by 1939.

Four words

Efference, afference, reafference, exafference

The 1950 paper is remembered for its vocabulary, and the vocabulary rewards precision. Efference is the outgoing motor command. Afference is the incoming sensory signal. Reafference is the portion of the incoming signal caused by the organism's own movement. Exafference is the portion caused by the world. The efference copy is the internal duplicate of the outgoing command, held back and compared against what returns. Reafference that matches the copy is canceled. Exafference survives the comparison and becomes news.

This is why the distinction is not academic. Reafference and exafference can be physically identical at the receptor: the same retinal slip, the same change in joint angle, the same stretch on the same spindle. Nothing in the signal itself declares who caused it. The only thing separating them is whether the system had ordered the movement that produced them. Authorship is not read off the sensation. It is computed by comparison, and the comparison is the entire mechanism. Strip the comparator out and the same nervous system, with the same receptors and the same muscles, loses the ability to tell its own action from an accident.

Eristalis

A fly with its head turned 180 degrees will circle without stopping

The experiment that carried the argument used a fly. Eristalis has a slender, flexible neck that can be rotated through 180 degrees about its long axis, and the rotated head can be fixed there with a spot of wax. Place a normal fly inside a vertically striped cylinder and rotate the cylinder, and the fly turns with it. That is the optomotor response, an old and reliable reflex: drifting stripes across the retina mean the animal has turned, so the animal turns back. Mittelstaedt was already running this preparation in 1949 (Mittelstaedt 1949), before the joint paper appeared.

Now invert the head. The retina is upside down, so a turn to the left delivers the retinal flow of a turn to the right. The correction the fly makes now runs in the same direction as the error. Negative feedback has become positive feedback. The animal circles, and the circling feeds itself. In darkness the same fly walks normally, and that control is what makes the result mean something: the machinery is intact, the sign of the loop is wrong. Popular retellings have the fly spinning until it collapses. The published finding is plainer and stronger. The sign of one comparison decides whether a body settles or runs away. One flip of that sign turns a stable animal into an unstable one without damaging a single part of it.

The older question

The problem was already more than a century old

Charles Bell described a version of the problem in 1823 (Bell 1823) and Johannes Purkinje in 1825. Hermann von Helmholtz took it furthest in the Handbuch der physiologischen Optik of 1867 (Helmholtz 1867), working in part from patients with paralyzed eye muscles. He called the compensating factor Willensanstrengung, the effort of will, and argued that the visual centers are informed of the intended change of gaze. His demonstration takes two seconds and needs no equipment. Move your eye and the room holds still. Press the side of your eyeball with a fingertip and the room jumps.

The gap between those two cases is the whole subject. The retinal displacement is comparable. What differs is whether a command was issued. Charles Sherrington offered the rival account in 1918 (Sherrington 1918): proprioceptors in the eye muscles could report position directly, with no internal copy required. Both channels turn out to exist, and the argument between them ran for decades. But Helmholtz had the shape of the answer 83 years before von Holst and Mittelstaedt gave it a mechanism, a diagram, and a name. Eighty three years is a long time to hold a correct intuition with no model to carry it.

Sperry

Two laboratories reached the same answer in the same year

Roger Wolcott Sperry published Neural basis of the spontaneous optokinetic response produced by visual inversion in the Journal of Comparative and Physiological Psychology, volume 43, pages 482 to 489, in 1950 (Sperry 1950). He had surgically rotated the eye of a fish and watched the animal circle. Same logic, different animal, different continent, same year. Sperry called the internal signal corollary discharge. Von Holst and Mittelstaedt called it efference copy. Sperry, born in 1913 and died in 1994, went on to share the 1981 Nobel Prize in Physiology or Medicine for the split brain work at Caltech (Sperry 1981).

The two terms get used interchangeably and often should not be. Efference copy, in the original formulation, is a copy of the motor command itself. Corollary discharge is the broader term for any motor derived signal sent into sensory structures, and it frequently names a gating or inhibitory action rather than a full replica. Careful writers keep them apart. That two independent groups converged inside a single year is the strongest evidence that the problem was ripe and the answer was forced by the data rather than invented. Sperry worked with fish, von Holst and Mittelstaedt with insects. The same architecture surfaced in both, which is a fair hint about how old it is.

The comparator

Silence is the sign that the prediction was correct

Put the pieces together and you have a control loop with a subtraction in it. The command goes out. The copy is held. The sensation returns. The two are compared and only the residue moves forward. When the body predicts itself accurately, the return is canceled and the system reports nothing. This is why you cannot reliably tickle yourself, why the visual world does not smear when you turn your head, why your own voice sounds quieter to you than it does to the room. Accurate self prediction shows up as absence.

Ask what that commits you to. If cancellation is the normal state, then most of what the nervous system does is invisible by design, and the only thing that reaches the surface is error. A quiet body is not a body with nothing happening in it. It is a body whose predictions are landing. That reframes sensation itself. What you notice is not a readout of what is present. It is a readout of what was not expected. That is a strong claim, and it is the claim the 1950 paper makes.

Von Holst and the model

Efference copy is self registration made experimental

Von Holst and Mittelstaedt turned self registration into an experiment. The Unified Model of Tone holds that what the body predicts, beneath everything else, is itself, and that the variable it predicts is tone. The 1950 paper is the first clean demonstration that a nervous system does this at all. It issues a command, holds a copy, predicts what the command will feel like, and forwards only the remainder. The world holds still when the eyes move because the prediction was correct.

Read through the reafference principle, tone is the standing expectation the nervous system holds about its own body: how much length, how much load, how much return each movement ought to produce. That makes tone a prediction rather than a property. It can be accurate or out of date independently of tissue integrity. A body can be structurally sound and still be defending a forecast written for conditions that no longer hold. The mismatch shows up as guarding, bracing, altered timing, and amplitude spent on a problem already solved.

The model states the same architecture in computational terms. Under active inference, descending motor commands are themselves proprioceptive predictions, and spinal reflex arcs function to minimize the error between predicted and actual body state. Movement, on that reading, is the nervous system acting to fulfill its own predictions about where the body should be. The comparator von Holst drew is not one stage inside movement. It is what movement is for. Karl Friston gave that loop its formal statement four decades later, and the model is careful to say that neither man proposed tone.

Then the claim this page carries furthest. The body cannot form an accurate model of where it is by holding still. It must move to sample itself, and each movement updates the map from which the next movement is planned. Engineers rediscovered the same principle when they built machines that have to locate themselves in space. An autonomous robot performs simultaneous localization and mapping, moving continuously while its sensors update both its model of the world and its estimate of its own position within that world. The motion is not separate from the mapping. The motion is the mapping, and a robot that cannot move cannot localize itself even with perfect sensors.

A 3D printer probes its bed point by point before it prints, because it can learn where the surface actually is only by touching it. An aircraft inertial navigation system runs known motions on startup to find true vertical and true heading, because orientation is knowable only by moving through it. The living body is the richest such system there is. It localizes itself by moving, and it cannot localize itself any other way. A body that stops moving stops knowing where it is. Say the boundary plainly. Von Holst and Mittelstaedt worked on flies, fishes, and eye movements, and they wrote nothing about tone. The comparator is theirs. The extension is the model's.

What survived

The 1950 model was too simple and it still won

Nobody now holds that the efference copy is a literal duplicate of the motor command. Work on smooth pursuit treats it as a prediction of expected sensory consequences rather than a replica of the outgoing signal. Mark Latash has asked in print what exactly the copy is a copy of (Latash 2021). Anatol Feldman has argued that referent control accounts for the same phenomena with no copy at all, noting that alpha gamma coactivation during movement enhances reafference rather than suppressing it (Feldman 2016). These are real objections and they remain open. Naming them is not a weakness of the model. It is how a hypothesis that began with a fly in a striped drum is still being argued over seventy five years later.

What survived is the architecture. Forward models in motor control, predictive coding in perception, and active inference in theoretical neuroscience are all versions of the same loop: predict, compare, act on the difference. Von Holst's 1954 English restatement in the British Journal of Animal Behaviour, volume 2, pages 89 to 94, has been cited more than a thousand times (von Holst 1954). Here is what this page adds to the story of tone: the reafference principle is the reason a body can be entirely intact and still be wrong about itself, because the nervous system does not store the body, it stores a forecast of the body. The same logic runs forward into the physics of the nerve impulse and backward into every model of posture that came before it.

What the record shows

Von Holst built the predictive brain out of eels, flies, and coupled rhythms

• 1950. Das Reafferenzprinzip appeared in Naturwissenschaften, volume 37, issue 20, pages 464 to 476 (von Holst and Mittelstaedt 1950). Thirteen pages, and they changed what a feedback loop was allowed to mean.
• 1939. Von Holst named the maintenance tendency and the magnet effect in fishes, describing coupled oscillator behavior in a fish tank a generation before it became the formal core of movement science (von Holst 1939).
• The eel, through the 1930s. A deafferented eel still swam, the wave traveling down the body in order and at rate with no sensory signal available to trigger the next segment. Von Holst summarized the fish work in 1937 (von Holst 1937), and the result cut directly against the reflex chain.
• Eristalis, head rotated 180 degrees. With the retina inverted, negative feedback became positive feedback and the fly circled without stopping. In darkness the same fly walked normally. Mittelstaedt was running the preparation in 1949 (Mittelstaedt 1949).
• 1867. Hermann von Helmholtz had already located the problem in the Handbuch der physiologischen Optik, calling the compensating factor Willensanstrengung, the effort of will (Helmholtz 1867). Charles Bell described a version in 1823 and Johannes Purkinje in 1825.
• 1950, independently. Roger Sperry published on the optokinetic response produced by visual inversion in the Journal of Comparative and Physiological Psychology, pages 482 to 489, and named the same class of signal corollary discharge (Sperry 1950).
• 1954. The English restatement in the British Journal of Animal Behaviour, volume 2, pages 89 to 94, has been cited more than a thousand times (von Holst 1954).