High-resolution unsteady three-dimensional flow simulations in large intracranial arterial networks of a healthy subject and a patient with hydrocephalus have been performed. The large size of the computational domains requires the use of thousands of computer processors and solution of the flow equations with approximately one billion degrees of freedom. We have developed and implemented a two-level domain decomposition method, and a new type of outflow boundary condition to control flow rates at tens of terminal vessels of the arterial network. In this paper, we demonstrate the flow patterns in the normal and abnormal intracranial arterial networks using patient-specific data.