Most bilingual lexical models assume that L1 and L2 either share the same semantic system, or are distinguished at the semantic level but connected through lexical associations. For example, the Revised Hierarchical Model (Kroll & Stewart, 1994) assumes the stronger access from L2 to concepts via the L1 lexical representation at the early stage of L2 acquisition and direct access to concepts after L2 proficiency is achieved. However, the model is not well supported by subsequent empirical evidence, and encounters difficulty in explaining cross-language priming data. The recently developed Sense Model (Finkbeiner, M., Forster, K., Nicol, J., & Nakamura, K., 2004) assumes a direct access from the L2 form to its related meaning and argues for the representational asymmetry in lexical semantics between L1 and L2. This model was designed to account for the translation asymmetry and task effect in the masked priming literature: L2-L1 priming is not observed in lexical decision due to the small proportion of L1 senses activated by the L2 prime; however, the category provides a context which restricts L1 sense activation and thus enhances the effectiveness of the L2 prime in semantic categorization. This dissertation reports the results of several semantic categorization experiments designed to test several assumptions of the Sense Model. Experiments 1-4 investigated whether the Category Restriction Hypothesis assumed by the Sense Model was empirically supported when congruence effects are minimized. The results showed that translation priming could be obtained for exemplars when congruence effects were controlled, but that there were no effects for non-exemplars, as predicted by the Sense Model. Subsequent experiments showed that category size is an important variable, since L2-L1 priming was not obtained with large categories (e.g., living thing), which was taken to indicate that a large category is ineffective as a 'focusing' device to enhance the activation of L2 semantic senses. Finally, it was shown that the priming asymmetry in lexical decision was not due to differential degrees of semantic activation of the prime in L1 and L2.