We show that second-language perception can be problematic if an L2 phoneme has multiple correspondents in the native language. As an example, we tested the perception of Spanish vowels by Dutch learners of Spanish. It turns out that the average learner indeed shows poorer identification performance for short front vowels (of which Spanish has two and Dutch has three) than for short back vowels (of which both languages have two), which is the exact reverse of what L1 Spanish listeners do. The learners have different perception strategies according to the language that they think they hear: when having to identify the vowels in Spanish CVC contexts by using Dutch vowel categories, they use the extra L1 category (/I/) much less than if they think that the language is Spanish rather than Dutch. This difference between the two modes of perception turns out to correlate strongly with the subjects' performance on an identification task with Spanish response categories. This suggests that learners try to solve the problem of the extra category by gradually removing it from their perception.