One of the main reasons for the popularity of neural networks lies in the ability to learn from data and/or experiences, which may compensate for what most expert systems might lack. However, since data may be incomplete and/or inaccurate and situations are usually uncertain and to some extent different from experiences, crisp decisions based on data or experiences are apt to fail in new situations. Therefore, the application of probabilistic inference seems indispensable in some circumstances. A probabilistic Hamming net is designed for discrete variables to effectively compute the probabilities and to provide explicit matching and multiplexing mechanisms.