Learning a phonetic category (or any linguistic category) requires integrating different sources of information. A crucial unsolved problem for phonetic learning is how this integration occurs: how can we update our previous knowledge about a phonetic category as we hear new exemplars of the category? One model of learning is Bayesian Inference, which gives a mathematical account of how previous knowledge (the "prior") interacts with new evidence coming in (the "likelihood"). Bayesian inference has been successfully applied in explaining word learning (Xu & Tenenbaum, 2007). This study asks whether a Bayesian model can account for the integration of statistical information in phonetic learning, investigating the process of learning two lexical tones (high vs. low) by English speakers. Four experiments explore whether subjects are sensitive to the two aspects of Bayesian models: prior knowledge of the tone categories (e.g., whether a category is rare or frequent) and category likelihood (whether the category has a wide or narrow variance in values). Experiments 1 and 2 investigated whether learners are sensitive to prior information about how frequently or rarely the two tone categories occur. Subjects, after completing a pitch discrimination task, were trained to associate five low tones with the color pink and five high tones with blue via a picture-sound association task. Subjects received either biased training with a 5:1 ratio between the two color categories or unbiased training with a 1:1 ratio. Subjects then performed an identification post-test, including all ten tone exemplars. The results show that poor pitch discriminators categorized significantly more sounds into the biased category, while good pitch discriminators were not biased by the high-frequency category. The same results were found in the generalization of the tonal contrast to novel tokens. These results suggest that prior knowledge is indeed used in phonetic learning,and also show individual differences in how the prior is incorporated. Experiments 2 and 3 explored the role of distributional variability within phonetic categories in learning. Subjects received training with either a small variance or large variance in the realization of tones. The results show that subjects learned significantly better when trained with large-variance input, regardless of their auditory sensitivity. Thus, exposure to within-category variation facilitates the learning of a highly variable phonetic category. The same results were found in the generalization of the contrast to novel tokens. The findings suggest that the category likelihood specified in the Bayesian model does affect learning. In summary, the study demonstrates the potential use of the Bayesian model in understanding the influence of statistical information in phonetic category learning. Moreover, the study support the probabilistic nature of phonetic category learning, and extends previous work such as Maye, Weiss & Aslin (2008) on the role of distributional learning by showing both how the influence of statistical cues fits into a Bayesian framework and the key need for modeling individual differences in using distributional cues. [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://www.proquest.com/en-US/products/dissertations/individuals.shtml.]