A potential high-level radioactive waste (HLRW) disposal site in northwestern China was investigated to determine its suitability for such a use. The site is primarily covered with welldeveloped metamorphic granite rocks. The primary targets for geological repositories are three granite rock masses (I, II, and III). Only surface geological data were available from previous studies. The objective of this study was to evaluate the quality of the rock mass and identify any weak geological structures that could jeopardize this future underground repository. We used gravity and aeromagnetic data on a large scale to study the regional geological structures within and around the three rock masses. Subsequently, in 2009, a controlled source audio-frequency magnetotelluric (CSAMT) survey was conducted to study rock mass I in more detail.This paper introduces three-dimensional (3-D) tomography imaging of gravity and aeromagnetic data. 3-D tomography imaging was carried out on previously collected gravity and aeromagnetic data and, using the results of different depth slices, we evaluated the rock mass quality and interpreted the geology. The aeromagnetic depth slices show that at about 1-km deep in rock masses I and III there is a high magnetic susceptibility body, possibly caused by an early granite intrusion. The deep granite or metamorphic rocks appear as weak`magnetic anomalies in the magnetic slices. The results also indicate that rock mass II is more deeply buried and smaller than rock masses I and III. In the upper crust, inhomogeneous density variations have a strike direction consistent with that of the structure revealed in the aeromagnetic data.For rock mass I, we designed a CSAMT line to map small-scale outcropped faults to depth, and identify any hidden faults or weakened zones in the subsurface. Because of the highly resistive ground, we improved the electrode contact conditions by pouring salt water on both the transmitting and receiving electrodes. During data processing, we applied Hanning filtering to reduce static effects from lack of electrical homogeneity near the ground surface, and through experimentation found an optimum filter length to achieve maximum static effect reduction. Sounding data were inverted for geoelectrical cross sections using a two-dimensional smoothmodel inversion method. Inversion artifacts were suppressed by imposing the model smoothness constraint. Several important results were revealed by the inversion: the inversion section correctly recognized two fractures at the surface, and also identified two new faults that were not observed during the previous geological survey. The inversion profile suggests that the narrow factures and deformation bands extend to a great depth. With further drilling confirmation, this interpretation will enhance understanding of the deep geological structures at the HLRW disposal site.