A major requirement for theory development in L2 vocabulary acquisition is an adequate model of L2 lexical organization and processing. At present, there is no widely accepted model of this kind, but recent research has demonstrated the potential usefulness of Levelt's (1989, 1993) L1 speech processing model (de Bot & Schreuder, 1993; Green, 1993; Poulisse, 1993). This paper explores the relevance of the model as adapted for L2 lexical processing of written texts, with particular attention to its mental lexicon components (concepts, lemmas, and lexemes) and organization. The model is first discussed in terms of its relevance to L2 lexical access, retrieval, and acquisition processes. Findings of an introspective L2 vocabulary study follow, detailing how university ESL learners reported dealing with unfamiliar words while carrying out L2 reading tasks. The information provided on their lexical inferencing strategy and varied knowledge sources used in the process provide support for the lexical components and organization proposed by Levelt. It is argued in this paper that the process of inferring the features of unknown words in a reading passage can be described in terms of lemma construction, as defined in the model.