Perception of English /r/ and /l/ is a well‐known difficulty for Japanese learning English. An identification test for minimal pairs read by two American speakers was administered to 32 Japanese students of English, first just after arrival in the United States, and then nine weeks later at the completion of an intensive English program emphasizing oral skills. Little improvement was observed after oral training. The perception of /r/ and /l/ as singletons and in consonant clusters exhibited quite opposite trends. In clusters, /l/ was perceived more accurately than /r/ (66% versus 52%), while for singletons, /l/ was somewhat worse than /r/ (64% versus 70%). Singleton consonants in word‐final position were more accurately perceived than initial singletons (77% versus 57%) while for clusters, the finals were slightly worse than initials (56% versus 62%). Thus both the /r‐l/ effect and word‐position effect interact with the singleton‐cluster factor but not with each other. These findings will be discussed in relation to the phonotactics of both Japanese and English and in terms of their acoustic correlates. [Supported by NIH, HD12511.]