The paper is devoted to graph theoretical analysis of functional connectivity matrices of 96 patients diagnosed with obsessive-compulsive disorder (OCD) and 95 healthy controls. Connectivity matrices were obtained from electroencephalography (EEG) as a solution of inverse problem, each matrix contains functional connectivity of 84 regions of interest (ROIs) corresponding to segmentation according to the Brodmann areas. The connectivity matrix was converted into a graph by applying a threshold. Small-network characteristics such as clustering coefficient and characteristic path length were computed as a function of threshold for 4 frequency bands. It was shown that the level of clustering coefficient is significantly lower for the OCD group than for healthy controls, which means loss in local integration. Characteristic path length for OCD patients was significantly higher than for controls, which indicates loss in multi-scale connections. Regression analysis was used for understanding connection between topological features of the network and severity symptoms evaluated by the Yale-Brown Obsessive-Compulsive scale (Y-BOC). Direct relation was demonstrated between the severity of the disease and local integration properties in theta band (corelation coefficient 0.1870, p-value 0.018208).