Lateral voltage spread in electrically coupled retinal horizontal cell networks is the substrate of center-surround antagonism in bipolar and ganglion cells. We studied its spatial and temporal properties in more detail in turtle L1 horizontal cells by using a contrast border as light stimulus. Experimental data were contrasted with expectations from a linear continuum model to specify the impact of nonlinearities. The assumptions for the diffusion term of the continuum model were justified by neurobiotin labeling. Measured voltage spread revealed two different length constants lambda+ and lambda0, under illuminated and nonilluminated regions of the retina, respectively, as predicted by the linear model. Length constants in the illuminated region showed strong temporal dynamics. For the initial phase of the horizontal cell responses lambda+ was larger than lambda0. This was also in accordance with the model. Right at the peak of the response, however, lambda+ dropped below lambda0 and did not change any more. It is this temporal reversal of asymmetry in voltage spread and not the decrease of lambda+ itself that is lacked by the linear model. The observed independence of the mean ratio lambda+/lambda0 from light intensity in both the peak and the plateau phases of horizontal cell responses contradicts the linear assumption, too. These two effects have to be addressed to local nonlinearities in the horizontal cell network like a negative feedback loop from photoreceptors and/or voltage-dependent conductances. Due to the failure of the linear model, firm conclusions about the membrane resistance and the coupling resistance of the horizontal cell network cannot be drawn from length constant measurements.