A computer simulation program for visual “pattern recognition” has been tested on an array of extremely degraded spoken digits. The program develops its own set of operators, as a function of its experience with inputs presented to it; so it does not depend upon preanalysis of a specific problem. The program gave (on an array of five different numbers) 100% success on words that had been presented to it for “training” purposes, and 100% success on words spoken by speakers different from those used in training. Results with alphabetic patterns, in which 5‐, 10‐, and 26‐letter arrays were all learned to 100% success, suggest that larger arrays of spoken words may give similar results.