This study examined the effects of the first language (L1) transfer and the second language (L2) proficiency on the perception and production of onset clusters by Korean speakers of English. Two cluster types (C+glide vs. C+liquid), the role of L2 proficiency (high, 19; middle, 18; low, 18), and two speech tasks (perception and production) were tested. To investigate the impact of the two above mentioned factors, two theoretical approaches were used: the Speech Learning Model (SLM) and the Perceptual Assimilation Model (PAM). The hypotheses derived from the two models were explored. As suggested by the results of the analysis, the PAM was found to have the strongest explanatory power in the perception task, while the SLM was in the production of onset clusters. Unequivocally, the biggest challenge for Korean learners in the perception task, regardless of their L2 proficiency are the C+liquid clusters. Whereas in the production task, the accuracy of C+liquid grew with an increase of L2 proficiency level; however, no such pattern was observed for the C+glide cluster.