When acquiring a second language, learners must become able to perceive the difference between novel sounds ("discrimination") and also associate strings of sounds with distinct vocabulary items ("lexical encoding"). Recent research has questioned whether discrimination is a sufficient or necessary condition for encoding, and Japanese length is of particular relevance to this question because of the large number of contrasts distinguished primarily by duration. For example, "koko," "kooko," "kokko," "kokoo," "kookoo," and "kokkoo" are six distinct, unrelated nouns, and while previous studies have shown that learners are able to detect differences between at least some forms, length-related errors in word recognition and production remain highly frequent, even for advanced learners. Therefore, modeling these error patterns is of key interest to teachers and researchers, but current models of L2 perception do not yet account for phonemic length. This study examines the performance of 30 American English-speaking learners of Japanese and 10 native speakers on an Oddity discrimination task, two lexical encoding tasks (picture-word matching and forced-choice lexical decision), and a Free Classification and Identification task to model patterns of confusion across a variety of length-related contrasts. For discrimination, vowel contrasts in initial position were easier than consonant length contrasts, and Free Classification and Identification results patterned very closely with discrimination performance, showing that these tasks are viable for modeling discriminability, even for non-segmental features. For lexical encoding, the most difficult contrasts in discrimination were similarly at near chance levels for lexical encoding, suggesting that a baseline level of discrimination is needed. Overall, there were moderate, significant correlations between discrimination and lexical encoding tasks, but the encoding tasks were much more closely related to each other than to discrimination, providing convergent and divergent evidence that the two constructs are measurably separable. For encoding, non-word competitors of the "all-short" form were much more difficult to reject compared to stimuli that had spurious long segments. These results line up with previous research on other languages to show how lexical encoding is hierarchical. In sum, the results support the interpretation that lexical encoding, while related to discrimination, is a separable construct. [The dissertation citations contained here are published with the permission of ProQuest LLC. Further reproduction is prohibited without permission. Copies of dissertations may be obtained by Telephone (800) 1-800-521-0600. Web page: http://bibliotheek.ehb.be:2222/en-US/products/dissertations/individuals.shtml.]