Why do second language learners fail to acquire certain L2 phonemic contrasts even long after they have "mastered" the L2 syntax? In this paper, I seek to provide an answer to this question. Research by Brown (1993) indicates that the learner's L1 feature geometry plays a role in the acquisition of L2 segmental phonology. This hypothesis is tested by examining the inability of Japanese, German, Turkish and French (Canadian and European) learners of English to correctly perceive /θ/ and /ð/. I argue that the lack of the feature [distributed] in these languages is the source of these errors. This feature marks the contrast between /θ-ð/ and /s-z/ in English. Although feature geometry cannot predict the specific errors for each language, I argue that it can constrain the list of possible candidates which will be substituted for the interdentals. The error phones perceived share the same structure or have minimally less structure than that of the target phone. I suggest that the specific error phone is then determined at the phonetic level of the learner's L1.