Purpose: This study investigated the efficacy of ultrasound biofeedback compared to a traditional acoustic input-only method in the second-language (L2) production training of English tense-lax vowel contrasts (/i-[near-close near-front unrounded vowel]/ and /e-[open-mid front unrounded vowel]/). The relationship between the degrees of production and perception learning was also examined to explore a potential L2 production-perception link. Method: Using a pretest/training/posttest paradigm, two groups of native Mandarin speakers received three sessions of training, one with ultrasound and one with audio only. Perception performance was measured by accuracy in AX discrimination. Production performance was measured in terms of accuracy in repetition production and Euclidean distance between contrasting vowels. Results: Both groups showed significant improvements in posttest production and perception tasks. Additionally, the training benefits were retained 6 weeks after training. The two groups, however, did not differ significantly in training gains; nor was there a significant correlation between training-induced changes in perception and production performance. Conclusions: The results indicate that ultrasound feedback is similarly effective as the audio-only training in improving the L2 production of English tense-lax contrasts and the benefits transferred to perception. The lack of correlation between production and perception gains suggests that the two modalities need to be separately trained to achieve desired L2 learning outcomes.