We investigated by a computational model of the basal ganglia the different network effects of deep brain stimulation (DBS) for Parkinson's disease (PD) in different target sites in the subthalamic nucleus (STN), the globus pallidus pars interna (GPi), and the globus pallidus pars externa (GPe). A cellular-based model of the basal ganglia system (BGS), based on the model proposed by Rubin and Terman (J Comput Neurosci 16:211-235, 2004), was developed. The original Rubin and Terman model was able to reproduce both the physiological and pathological activities of STN, GPi, GPe and thalamo-cortical (TC) relay cells. In the present study, we introduced a representation of the direct pathway of the BGS, allowing a more complete framework to simulate DBS and to interpret its network effects in the BGS. Our results suggest that DBS in the STN could functionally restore the TC relay activity, while DBS in the GPe and in the GPi could functionally over-activate and inhibit it, respectively. Our results are consistent with the experimental and the clinical evidences on the network effects of DBS.