This dissertation provides a new model of the phonology-morphology interface, focusing on Phonologically Conditioned Allomorphy (PCA). In this model, UR selection occurs during the phonological component, and mappings between meanings and URs are encoded as violable constraints, called UR constraints (Boersma 2001; Pater et al. 2012). Ranking UR constraints captures many empirical generalizations about PCA, such as similarities between PCA and phonological alternations, the existence of defaults, and the interaction of PCA and phonological repairs (epenthesis, deletion, etc.). Since PCA follows from the ranking or weighting of constraints, patterns of PCA can be learned using existing learning algorithms, and modeling variation in PCA is straightforward. The main empirical conclusion is that some cases of PCA are driven by the general phonological grammar, and are not the result of subcategorization. This conclusion follows from three case studies: English a and an, French liaison, and English suffixes -(a)licious and -(a)thon. For each of these cases, PCA is closely integrated into the phonology of the language. The same ranking of constraints can be used to capture both alternations and PCA, and phonological repairs like epenthesis and deletion conspire with PCA to avoid marked structures. Each of these cases comes with novel data and analyses. For English a and an, the selection of the article interacts with h-deletion and ʔ-epenthesis, conspiring with both to avoid hiatus. In French liaison, the default allomorph is often unpredictable, and liaison in some words conspires with n-epenthesis to avoid hiatus. For English -(a)licious and -(a)thon, PCA avoids hiatus and stress clash, and the suffix -(a)licious conspires with the Rhythm Rule to avoid stress clashes. For each case study, I also present variable data, analyzed with UR constraints in Maximum Entropy Harmonic Grammar (Goldwater and Johnson 2003).