This paper presents the results of a series of experiments in acoustic phonetic feature-based recognition of phonemic segments from continuous speech. The work is directed toward the development of a computationally efficient framework which combines acoustic phonetic knowledge and statistical classification techniques. The statistical techniques selected are multivariate Gaussian classifiers which are applied to the problem of identifying the 20 vocalic phonemes (12 vowels and 8 diphthongs) of British Received Pronunciation. Current results of classification of hand segmented data show cumulative correct-in-rank scores of 71%, 88%, and 94% in the top 3 ranks for one male talker and 66%, 85%, and 90% for a second male speaker on test sets of 98 sentence-length utterances containing a total of approximately 750 vowels.