Obsessive-compulsive disorder (OCD) is a clinically heterogeneous disorder characterized by multiple, temporally stable symptom dimensions. Preliminary functional neuroimaging studies suggest that these symptom dimensions may have distinct neural substrates. Whole-brain voxel-based morphometry was used to examine the common and distinct neuroanatomical (structural) substrates of the major symptom dimensions of OCD. First, we compared 55 medication-free patients with OCD and 50 age-matched healthy control subjects. Multiple regression analyses were then used to examine the relationship between global and regional grey matter (GM) and white matter (WM) volumes and symptom dimension scores within the patient group. OCD patients showed decreased GM volume in left lateral orbitofrontal (BA47), left inferior frontal (BA44/45), left dorsolateral prefrontal (BA9) and right medial prefrontal (BA10) cortices and decreased bilateral prefrontal WM volume. Scores on the 'symmetry/ordering' dimension were negatively correlated with 'global' GM and WM volumes. Scores on the 'contamination/washing' dimension were negatively correlated with 'regional' GM volume in bilateral caudate nucleus and WM volume in right parietal region. Scores on the 'harm/checking' dimension were negatively correlated with regional GM and WM volume in bilateral temporal lobes. Scores on the 'symmetry/ordering' dimension were negatively correlated with regional GM volume in right motor cortex, left insula and left parietal cortex and positively correlated with bilateral temporal GM and WM volume. The results remained significant after controlling for age, sex, educational level, overall illness severity, global WM and GM volumes and excluding patients with comorbid depression. The reported symptom dimension-specific GM and WM alterations support the hypothesis that OCD is an etiologically heterogeneous disorder, with both overlapping and distinct neural correlates across symptom dimensions. These results have clear implications for the current neuroanatomical model of OCD and call for a substantial revision of such model which takes into account the heterogeneity of the disorder.