Teuvo Kohonen has recently developed an algorithm similar to that used in his feature map classifiers but in which learning is supervised rather than unsupervised. This algorithm, known as learning vector quantization (LVQ), is similar to a K‐nearest neighbor algorithm and allows a system to learn the vector quantization of the inputs to different categories. This algorithm is very simple, does not require a large number of training trials, and is capable of forming complex decision regions. As a recognition task, the speaker‐dependent recognition of the phonemes /b/, /d/, and /g/ in different phonetic contexts is considered. The training procedure is applied to speech patterns that are stepped through in time, thus providing the system with a measure of shift invariance. Preliminary results indicate that LVQ can yield a recognition rate of 98.3% for 1880 testing tokens from three speakers. The simple vector operations that constitute the core of LVQ allowed for very easy parallelization and thus high learning speed, i.e., less than an hour.