LETTER Communicated by Michael Berry Rate Coding Versus Temporal Order Coding: What the Retinal Ganglion Cells Tell the Visual Cortex Rufin Van Rullen Simon J. Thorpe Centre de Recherche Cerveau et Cognition, Facult´ e de M´ edecine Rangueil, 31062 Toulouse Cedex, France It is often supposed that the messages sent to the visual cortex by the retinal ganglion cells are encoded by the mean firing rates observed on spike trains generated with a Poisson process. Using an information trans- mission approach, we evaluate the performances of two such codes, one based on the spike count and the other on the mean interspike interval, and compare the results with a rank order code, where the first ganglion cells to emit a spike are given a maximal weight. Our results show that the rate codes are far from optimal for fast information transmission and that the temporal structure of the spike train can be efficiently used to maximize the information transfer rate under conditions where each cell needs to fire only one spike. 1 Introduction How do neurons transmit information? This question is a central problem in the field of neuroscience (Perkel & Bullock, 1968). Signals can be conveyed by analog and electrical mechanisms locally, but over distances information has to be encoded in the spatiotemporal pattern of trains of action potentials generated by a population of neurons. The exact features of these spike trains that carry information between neurons need to be defined. The most commonly used code is one based on the firing rates of individual cells, but this is by no means the only option. In recent years a strong debate has opposed partisans of codes embedded in the neurons’ mean firing rates and researchers in favor of temporal codes, where the precise temporal structure of the spike train is taken into account (Softky, 1995; Shadlen & Newsome, 1995, 1998; Gautrais & Thorpe, 1998). Here we address this question of neural coding in the context of information transmission between the retina and the visual cortex. The retina is a particularly interesting place to study neural information processing (Meister & Berry, 1999). First, it is relatively easy to stimulate and record retinal cells. Furthermore, the general architecture and functional organization of the retina are remarkably well known (Rodieck, 1998). There is probably no other place in the visual system where one can define more Neural Computation 13, 1255–1283 (2001) c  2001 Massachusetts Institute of Technology