A method and apparatus for defining the location of a medical instrument relative to features of a medical workspace including a patient's body region are described. Pairs of two-dimensional images are obtained, preferably by means of two video cameras making images of the workspace along different sightlines which intersect. A fiducial structure is positioned in the workspace for defining a three dimensional coordinate framework, and a calibration image pair is made. The calibration image pair comprises two 2D projections from different locations of the fiducial structure. After the calibration image pair is made, the fiducial structure is removed. A standard projection algorithm is used to reconstruct the 3D framework of the fiducial structure from the calibration image pair. Appropriate image pairs can then be used to locate and track any other feature such as a medical instrument, in the workspace, so long as the cameras remain fixed in their positions relative to the workspace. The computations are desirably performed with a computer workstation including computer graphics capability, image processing capability, and providing a real-time display of the workspace as imaged by the video cameras. Also, the 3D framework of the workspace can be aligned with the 3D framework of any selected volume scan, such as MRI, CT, or PET, so that the instrument can be localized and guided to a chosen feature. No guidance arc or other apparatus need be affixed to the patient to accomplish the tracking and guiding operations.