Here we present the first model that quantitatively predicts the apparent spatial position and shape of percepts elicited by retinal electrical stimulation in humans based on the known anatomy of the retina. This model successfully predicts both the shape of percepts elicited by single electrode stimulation and the shape and relative positions of percepts elicited by multiple electrode stimulation. Model fits to behavioral data show that sensitivity to electrical stimulation is not confined to the axon initial segment, but does fall off rapidly with the distance between stimulation and the initial segment. Using the model, it is possible to compensate, preferably with a look up table, to match percepts to a desired image.