A probabilistic method based on the Hamming distance is presented for calculating the pattern-recognition response of a class of single-layer microcircuit learning nets. The prediction of the behaviour of a 12-element machine is shown to agree with measured responses. The method is also used to explain the characteristics of nets trained on translated versions