The authors present the results of accuracy measurements, obtained in both laboratory phantom studies and an in vivo assessment, for a technique of frameless stereotaxy. An instrument holder was developed to facilitate stereotactic guidance and enable introduction of frameless methods to traditional frame-based procedures. The accuracy of frameless stereotaxy was assessed for images acquired using 0.5-tesla or 1.5-tesla magnetic resonance (MR) imaging or 2-mm axial, 3-mm axial, or 3-mm helical computerized tomography (CT) scanning. A clinical series is reported in which biopsy samples were obtained using a frameless stereotactic procedure, and the accuracy of these procedures was assessed using postoperative MR images and image fusion. The overall mean error of phantom frameless stereotaxy was found to be 1.3 mm (standard deviation [SD] 0.6 mm). The mean error for CT-directed frameless stereotaxy was 1.1 mm (SD 0.5 mm) and that for MR image-directed procedures was 1.4 mm (SD 0.7 mm). The CT-guided frameless stereotaxy was significantly more accurate than MR image-directed stereotaxy (p = 0.0001). In addition, 2-mm axial CT-guided stereotaxy was significantly more accurate than 3-mm axial CT-guided stereotaxy (p = 0.025). In the clinical series of 21 frameless stereotactically obtained biopsies, all specimens yielded the appropriate diagnosis and no complications ensued. Early postoperative MR images were obtained in 16 of these cases and displacement of the biopsy site from the intraoperative target was determined by fusion of pre- and postoperative image data sets. The mean in vivo linear error of frameless stereotactic biopsy sampling was 2.3 mm (SD 1.9 mm). The mean in vivo Euclidean error was 4.8 mm (SD 2 mm). The implications of these accuracy measurements and of error in stereotaxy are discussed.