Considering that L1 translation in L2 lexical encoding is commonplace in vocabulary learning in Korea, the study sought the type of L1 translation that can efficiently promote the target L2 processing. For this, 77 English majors at a university participated in two vocabulary learning conditions using ‘L1α’ (dictionary L1) and ‘L1β’ (individual learner’s own L1 reflecting his/her own speech style in actual communication) and were tested in a vocabulary production test reflecting L2 communication. Using a paired-samples t-test in a within-subjects design, the effects of the two L1 translation types were compared in the immediate and delayed posttests. The results suggested that the advantage of ‘L1β’ over ‘L1 α’ in L2 production was evident both for single-word and multi-word lexical items. The effect of ‘L1β’ was also confirmed in long-term memory. In the frame of computational models of lexical processing, the study suggested that the cognitive path ‘Concept-L1β-L2’ is more efficient than ‘Concept-L1β-L1α-L2’. Consequently, in the learning process, encoding new L2 lexical items through ‘L1 β’ rather than ‘L1α’ may be more effective to retrieve the target L2 in production. Pedagogical suggestions are discussed based on this model.