A three-dimensional (3-D) stochastic network model for simulating a hot dry rock (HDR) or hot wet rock (HWR) engineered geothermal system formed in fractured crystalline rock is presented. The model addresses the problems of fracture network characterization from in situ field data, such as fracture orientation, size, spacing, and other mechanical properties. The model can simulate the changes that occur within the rock mass during stimulation (i.e., large volume fluid injection at pressures sufficient to allow shear slip on natural fractures). It can also be used to simulate steady state circulation of the heat exchange system thus created and includes provision for predicting tracer response curves and heat extraction history. The model has been applied to data gathered during the stimulation and circulation of a 2.2-km-deep HDR reservoir at Hijiori, Japan. The predicted shape of the stimulated and shear-propped fractures closely matched the distribution of seismic source distribution of acoustic emission (AE), regardless of realization of the fracture network, suggesting that the geometry of the stimulated volume can be robustly predicted from knowledge of the fracture population and in situ stresses. However, hydraulic behavior and tracer tests during the circulation could only be satisfactorily simultaneously reproduced by a small subset of realizations. These selected realizations, obtained by matching initial circulation and tracer data, are considered to give the best prospect of satisfactory long-term thermal modeling. The success in simultaneously modeling diverse data (hydraulic, microseismic, and tracer) lends confidence to the thermal predictions. The results indicate that a large improvement in the long-term thermal performance of the Hijiori reservoir could result from increasing well spacing from 100 to 150 m without major degradation of the hydraulic performance. ¿ 2000 American Geophysical Union