In this talk I focus on second language learners’ representations of their target language sound system. I present data from two studies that suggest learner representations are i) finely tuned to contextual information in the input and ii) are gradient in nature. The studies examine the acquisition of the stop-approximant alternation ([b, d, g] ~ [beta, eth, g] in Spanish by L1 English/ L2 learners of Spanish. In Study 1 I consider how learners become sensitive to the conditioning factors of word position and stress that drive the alternation. In Study 2 I analyze learner productions of the same alternation. I integrate the findings into a theory of how learners build the representations required to acquire a second language sound system.