Abstract: A Bayesian blind equalizer which approximates the optimum symbol-by-symbol detector for an unknown intersymbol interference pattern in a communication channel is provided in a plurality of parallel processors. Each processor operates in parallel from a common data bus with each of the other processors. Each of the processors in turn generates an estimated signal and updated metric for the communication channel for a corresponding one of each of the possible data subsequences which could cause intersymbol interference. The estimated signals or innovations are then combined with the updated metrics in a supervisory processor to generate unconditional channel coefficients for the next received data sample. Using the estimated channel coefficients and received data samples, the transmitted data sample is reliably decoded notwithstanding intersymbol interference without the use of data preambles or training data and notwithstanding that the channel coefficients may be rapidly varying.