To evaluate the spatial accuracy of a rapid interactive method of transferring computed tomographic (CT) information between its display on a computer screen to its source (test object, operating field), a multidimensional computer combined with a six-jointed position-sensing mechanical arm was tested with a Plexiglas model consisting of 50 rods of varied height and known location, a plastic replica of the skull, and, subsequently, three patients. The median error value between image and real location was 1-2 mm (P > .95), regardless of the registration target sites. The accuracy, however, increased with the selection of widespread registration points, and 95% of all errors were below 3.70 mm (P > .95). The results compare favorably with the four most commonly used stereotaxic framed units. A misregistration error of 0.3-2.2 mm was found during intraoperative correlation between anatomy on the CT display and actual anatomic location in the operative field.