Image fusion, target localization, and setup accuracy of cone-beam computed tomography (CBCT) for stereotactic radiosurgery (SRS) were investigated in this study. A Rando head phantom rigidly attached to a stereotactic Brown-Roberts-Wells (BRW) frame was utilized to study the geometric accuracy of CBCT. Measurements of distances and angular separations between selected pairs of multiple radio-opaque targets embedded in the head phantom from a conventional simulation CT provided comparative data for geometric accuracy analysis. Localization accuracy of the CBCT scan was investigated from an analysis of BRW localization of four cylindrical objects (9 mm in diameter and 25 mm in length) independently computed from CBCT and conventional CT scans. Image fusion accuracy was quantitatively evaluated from BRW localization of multiple simulated targets from the CBCT and conventional CT scan. Finally, a CBCT setup procedure for stereotactic radiosurgery treatments was proposed and its accuracy was assessed using orthogonal target verification imaging. Our study showed that CBCT did not present any significant geometric distortions. Stereotactic coordinates of the four cylindrical objects as determined from the CBCT differed from those determined from the conventional CT on average by 0.30 mm with a standard deviation (SD) of 0.09 mm. The mean image registration accuracy of CBCT with conventional CT was 0.28 mm (SD = 0.10 mm). Setup uncertainty of our proposed CBCT setup procedure was on the same order as the conventional framed-based stereotactic systems reported in the literature (mean = 1.34 mm, SD = 0.33 mm). In conclusion, CBCT can be used to guide SRS treatment setup with accuracy comparable to the currently used frame-based stereotactic radiosurgery systems provided that intra-treatment patient motion is prevented.