Objective: To establish the impact of the imaging modality, registration method and use of intraoperative computed tomography (CT) scan on the accuracy of the ROSA® stereotactic robot. Methods: Using a dedicated phantom device, we measured the accuracy of the stereotactic robot for 20 targets as a function of the registration method (frameless, FL, or frame-based, FB) and the reference imaging modality (3T magnetic resonance imaging, MRI, CT scanner or flat-panel CT, fpCT). We performed a retrospective study of the accuracy of the first 26 FB and 31 FL robotized stereotactic surgeries performed in our department. Results: In a phantom study, the mean target accuracy was 1.59 mm for 3T MRI-guided FL surgery, 0.3 mm for fpCT-guided FL surgery and 0.3 mm for CT-guided FB surgery. In our retrospective series, the mean accuracy was 0.81 mm for FB stereotactic surgery, 1.22 mm for our 24 stereotactic surgery procedures with FL (surface recognition) registration and 0.7 mm for our 7 stereotactic surgery procedures with FL fiducial marker registration. Intraoperative fpCT fully corrected all the registration errors. Conclusions: The ROSA stereotactic robot is highly accurate. Robotized FB stereotactic surgery is more accurate than robotized FL stereotactic surgery.