A recognition system based on linguistic features was developed for the semivowels /w j r l/ in American English. The features of interest aresonorant,syllabic,consonantal,high,back,front, andretroflex. Acoustic correlates and events related to these features were used to detect and classify the semivowels. The recognizer was tested across semivowels occurring in a wide range of phonetic environments. The corpora included polysyllabic words and sentences spoken by males and females of several dialects. The results show that a feature‐based approach to recognition is a viable methodology. Fairly consistent overall recognition results were obtained. Across the test data, acoustic events were detected within 97% of the semivowels and classification rates were 62% for /w/, 74% for /l/ (/w/ and /l/ were often confused), 90% for /r/ and 84% for /j/.