Stereotactic angiography is the preferred imaging modality in the radiosurgical treatment of arteriovenous malformations (AVM). A major limitation of this technique is the inability to determine the three-dimensional shape and the volume of the AVM. We developed a technique for reconstructing the AVM from multiple stereotactic angiograms. The fiducial marks and the identified target area are digitized for each view. The 2-D target area is then stereotactically backprojected into a 3-D volume. The spatial boundary of the target volume is defined by superimposing the backprojections from each of the views. In feasibility studies with a phantom, this technique accurately reconstructed complex 3-D structures, depicting fine details as small as 1 mm. Clinical studies in two patients indicated that this novel technique is very useful for the 3-D reconstruction of AVM for radiosurgical treatment. The number of views required depends on the complexity of the object. In general, six to eight views appear adequate. The accuracy of the reconstruction can be compromised if the view angles are limited as with the current radiographic frame. A new frame design with practically unlimited view angles is proposed to avoid this potential limitation. However, structures that are always obscured (such as concavity) could not be accurately reconstructed, leading to overestimates of the object size. This "limitation" has a positive side, in that the technique will not underestimate the actual target volume. The ability to delineate both the shape and volume of the lesion should facilitate optimal target coverage while sparing a significant amount of normal tissue.