Two alternate methods for the selection of speech material to be used in studies in automatic speech recognition are suggested. Preference is given to the selection of a restricted set of phonemes from which words and sentences may be constructed for analysis. The selection can be based on the high frequency of occurrence of the individual phonemes, on the high yield of words by the phonemes, or on some combination of these two criteria. A combination of the two criteria appears to be most satisfactory. The phoneme selection was derived from the 500 most frequently occurring words, as given by Thorndike and Lorge. Some consideration was also given to the frequencies of occurrence of the phonemes. The phonemes are presented in cumulative order according to the number of words by which the list is increased as each successive phoneme is added.