Although obsessive-compulsive disorder (OCD) has long been a unitary diagnosis, there is much recent interest in its potential heterogeneity, as manifested by symptom subgroups. This study evaluated existing models of symptom structure in a sample of 203 individuals with OCD. Using confirmatory factor analysis, we examined the ability of each model to account for two levels of data: a priori symptom groupings (second-order) and individual symptoms, identified by the Yale-Brown Obsessive Compulsive Scale symptom checklist. Four models were examined: a single-factor, a two-factor (i.e., obsessions and compulsions), and two multidimensional models, comprising three and four factors. Adequate fit was found solely for the four-factor model--specifying obsessions/checking, symmetry/ordering, contamination/cleaning, and hoarding--but only at the second-order level; it did not account for relationships among discrete symptoms. Parameter estimates showed within-factor heterogeneity, as well as overlap between factors, most notably the two representing checking and contamination-related symptoms. The implications of these findings are discussed. Results provide evidence for the multidimensionality of OCD symptoms, but suggest that a comprehensive model has yet to be identified. They also point to the inadequacy of groupings based solely upon overt behavioural similarities (e.g., 'checking'). Recommendations are made for future research.