Three-dimensional (3D) volume reconstruction is easily feasible with axial, coronal or sagittal CT and MRI scans. On the other hand, radiosurgical treatment of arteriovenous malformations (AVMs) is exclusively based on two orthogonal stereotactic projections, obtained with angiographic procedures. Most commonly AVM volumes have been calculated by assimilating the nidus volume to a prolate ellipsoid. We present an algorithm dedicated to 3D volume reconstruction starting from two stereotactic orthogonal projections. This has been achieved using a euristic approach which has been widely adopted in the artificial intelligence domain. Tests on phantoms of different complexity have shown excellent results. The importance of the algorithm is considerable. Firstly, it allows calculations of complex structures far removed from a regular ellipsoid. Secondly, it permits shape recovery. Thirdly, it provides AVM visualization on axial planes.