Open shell CNDO/2 calculations have been performed with geometry optimization. The calculated geometries are in agreement with the symmetry as derived from ESR data. Phosphoranyl radicals are found to have trigonal bipyramidal geometries, with the exception of digand phosphoranyl radicals, which have tetrahedron-like geometries, when strong electron withdrawing ligands are absent. In a few calculations square pyramidal geometries were considered as deviations from ideal T or TBP structures. The spin densities, however, cannot be calculated accurately with CND0/2. A n-electron model is proposed to calculate spin densities in tetracoordinated n-ligand phosphorus complexes, in which the five 3d orbitals of phosphorus are included in the calculation. These spin densities are found to be in good agreement with ESR experiments. The success of this model, where only T electrons are explicitly calculated, is explained.