A method enabling a fully automated computer analysis of the visual field is described. Starting from the most probable assumptions about the sensitivity distribution within the visual field, this distribution is approximated in 4 steps. Decisions are based on probability theory. Every analytical step builds up on the conclusions reached in the preceding one. The theoretical limits of the method in regard to the degree of approximation of the true sensitivity function are discussed in detail. The spatial resolution is determined by the density of the questions per area of visual field. For a grid constant of 3°, the spatial resolution of sensitivity defects varies between 4 and 7°. Otherwise, the quality of the approximation varies within large limits and depends on prior knowledge of the expected sensitivity distribution and on the noise in the system. This includes the patient's threshold fluctuation and the reliability of the answers. The whole analytical program was tested on a large number of artificial visual fields contained in computer storage which were then tested by the main program. It is shown that even in the presence of large sensitivity fluctuations and a considerable fraction of erroneous answers by the patient the method is still able to extract the data which are essential from the clinical point of view.