This study examines the relationship between the production and perception of English /r/ and /l/ by native Japanese adults learning English in the United States. For some subjects, production of the contrast was more accurate than their perception of it, replicating and extending a previous finding reported by Goto (1971) in Japan. The difficulty in perception of the liquid contrast varied with its position in the word. Prevocalic /r/ and /l/ in consonant clusters yielded the greatest perceptual errors, while word-final liquids were accurately perceived. This pattern of errors is not predictable on the basis of contrastive phonological analysis, but might be the result of acoustic-phonetic factors. Implications for second language pedagogy are discussed.