Laboratory training experiment was conducted in order to examine the effect of acoustic and semantic contexts when learning second language phoneme perception. Fifty minimal pairs of English words contrasting in /r/ and /l/ were produced by native speakers of American English in three conditions; in isolation (WD), within semantically neutral carrier sentences (NS), and within semantically contextual carrier sentences (CS). Participants were native speakers of Japanese, and were divided into three groups; each was trained to identify /r/ and /l/ in one of above three conditions. In pretest, identification accuracy varied by condition in the order, NS<WD<CS, which replicated the previous study [Rothwell and Akahane-Yamada, J. Acoust. Soc. Am. 112, 2386 (2002)]. It was also shown that the group trained with CS stimuli improved the ability to identify CS stimuli from pretest to post-test, but not WD and NS stimuli. In contrast, the effect of training using WD or NS stimuli generalized to all the stimulus conditions. These results suggest that the perception training utilizing the auditory input, in which acoustic information is the only clue to identify phonemes, is effective in cultivation of aural comprehension. Implications for foreign language education will be discussed. [Work supported by TAO, Japan.]