In fast speech, certain segments pronounced in careful speech may be deleted. Rules of a generative phonology have been used to account for fast speech forms. An alternative approach is suggested which views fast speech deletions as merely limiting cases of segment reduction, under conditions of increased tempo and/or casualness. To complement this approach, generative phonology rules should be adjusted and stated as phonetic realization rules. All segment reduction/deletions resulting from stylistic variation in speech should be specified in descriptive terms rather than in classificatory terms. The segment is then not explicitly deleted but rather is assigned a zero length. The descriptiveness of realization rules is necessary because the differences found in the lengths of the segments are gradational rather than abrupt and the outputs of such rules are not further subject to phonological rules of the language. Accounting for speech variation resulting from assimilation by phonetic realization rules must involve a more radical reappraisal of the form of such rules than is required to account for segment deletion. High-frequency amplitude Mingograf tracings and spectrograms reflecting variation in speech rate are provided. (Author/JK)