Computer-based training can be effective in improving second language learners' perceptions and productions of segmental speech contrasts. However, because most previous studies have addressed specific theoretical problems in speech learning, an impact on pedagogy has hardly been felt. Research participants are commonly subjected to rigid research paradigms with strict schedules, so that it is unclear whether research results can be applied in pedagogical contexts. Moreover, a limited range of speech sounds have been examined, with a disproportionate number of studies considering only consonants. In this investigation, we applied training techniques from previous work in a pedagogically oriented approach in which participants had some control over lesson content and worked at a self-determined pace. Sixteen native Mandarin and Cantonese speakers received training on three English vowel contrasts over two months. The training stimuli consisted of synthetic and natural utterances, presented in a graded fashion. In comparison with an untrained control group, trainees showed improved perceptual performance, transferred their knowledge to new contexts, and maintained their improvement three months after training. These findings support the feasibility of computer-based, learner-centered programs for second language pronunciation instruction. (Contains 3 figures and 1 table.)