The recently defined Sparkle model for the quantum chemical prediction of geometries of lanthanum(III) and lutetium(III) complexes within AM1 (J. Phys. Chem. A 2006, 110, 5897) has been extended to PM3. As training sets, we used the same two groups, one for each lanthanide, of 15 high-crystallographic-quality (R factor < 0.05 angstrom) complexes as was previously chosen to parametrize Sparkle/AM1. Likewise, in the validation procedure, we used the same Sparkle/AM1 validation sets of 60 additional La(III) and 15 additional Lu(III) complexes. The Sparkle/PM3 unsigned mean errors for all interatomic distances between the metal ions and the ligand atoms of the first sphere of coordination proved to be random around the means of 0.068 angstrom for lanthanum(III) and 0.076 angstrom for lutetium(III), thus being comparable to the respective Sparkle/AM1 values of 0.078 and 0.075 angstrom. Furthermore, effective-core-potential ab initio calculations on smaller subsets of such complexes led to similar accuracies. Sparkle/PM3 and Sparkle/AM1 are therefore made available to the researcher who must decide which of the models to use based on considerations of the impact of either PM3 or AM1 on the description of the ligands and the consequence of such a choice on the properties of interest.