This paper addresses the problem of optimal geometrical design for tactile sensors. To this aim the particular case in which sensor-object contact information can be reduced to the form of a binary tactile image is considered. Possible criteria for the evaluation of space-invariant (orthogonal and hexagonal grids) and space-variant (retina-like and "multifovea") dispositions of sensing elements are discussed. Simulation studies are carried on to compare space-invariant configurations in terms of rate of success in pattern recognition, and to evaluate the efficiency of the multifovea disposition in the calculation of position and orientation of sample objects. Finally, results obtained by applying different pattern recognition algorithms to the analysis of pseudovisual images provided by a real sensorized static platform are presented.