This paper compares two approaches to syllable-counting allomorphy (SCA; Kager 1996), a type of phonologically conditioned suppletive allomorphy in which the distribution of allomorphs is determined by the syllable count of the stem. One approach to modeling this phenomenon is an output optimization approach in which phonological (P) constraints involved in foot parsing are ranked ahead of morphological (M) constraints in Optimality Theory (OT), such that the phonological constraints determine which allomorph is selected. This is one use of the 'P >> M' OT ranking schema proposed by McCarthy and Prince (1993a,b). Based on evidence from a cross-linguistic survey of SCA, this paper demonstrates that the P >> M approach is not sufficient to model this phenomenon, since in nearly half of the cases surveyed, the distribution of allomorphs is not phonologically optimizing. Instead, the author proposes that subcategorization be used to model SCA. Under a subcategorization approach, affixes contain specifications for stems to which they can attach, including both morphological and phonological specifications. The author argues that subcategorization is superior to output optimization in modeling SCA since the subcategorization analysis predicts and easily accounts for cases of phonologically non-optimizing SCA such as those revealed by the survey described in this paper.