Obsessive-compulsive disorder (OCD) is the tenth most disabling medical condition worldwide. Twin and family studies implicate a genetic etiology for this disorder, although specific genes have yet to be identified. Here, we present the first large-scale model-free linkage analysis of both extended and nuclear families using both 'broad' (definite and probable diagnoses) and 'narrow' (definite only) definitions of OCD. We conducted a genome-scan analysis of 219 families collected as part of the OCD Collaborative Genetics Study. Suggestive linkage signals were revealed by multipoint analysis on chromosomes 3q27-28 (P = 0.0003), 6q (P= 0.003), 7p (P = 0.001), 1q (P = 0.003), and 15q (P = 0.006). Using the 'broad' OCD definition, we observed the strongest evidence for linkage on chromosome 3q27-28. The maximum overall Kong and Cox LODall score (2.67) occurred at D3S1262 and D3S2398, and simulation based P-values for these two signals were 0.0003 and 0.0004, respectively, although for both signals, the simulation-based genome-wide significance levels were 0.055. Covariate-linkage analyses implicated a possible role of gene(s) on chromosome 1 in increasing the risk for an earlier onset form of OCD. We are currently pursuing fine mapping in the five regions giving suggestive signals, with a particular focus on 3q27-28. Given probable etiologic heterogeneity in OCD, mapping gene(s) involved in the disorder may be enhanced by replication studies, large-scale family-based linkage studies, and the application of novel statistical methods.