Measured speech waveform data was used as a basis for partitioning an utterance into segments and for classifying those segments. Mathematical classifications were used instead of the traditional phonemes or linguistic categories. This involved clustering methods applied to hyperspace points representing periodic samples of speech waveforms. The cluster centers, or hyperphonemes (HPs), were used to classify the sample points by the nearest- neighbor technique. Speech segments were formed by grouping adjacent points with the same classification. A dictionary of 54 different words from a single speaker was processed by this method. 216 utterances, representing four more repetitions by the same speaker each of the original 54 words, were similarly analyzed into strings of hyperphonemes and matched against the dictionary by heuristically developed formulas. 87% were correctly recognized, although almost no attempt was made to modify and improve the initial methods and parameters.