Magnesium parameters for use with the semiempirical AM1 method are developed using a specially designed genetic algorithm. Parametrization priorities included development of a robust parametrization capable of describing a wide range of properties in diverse chemical environments, with emphasis on structural features of biologically relevant systems, e.g., chlorophylls. Specifically, the test data set included a selection of the heats of formation, geometric properties, dipole moment, and ionization energies evaluated for 32 compounds including halides, oxides, hypervalent compounds, organometallics, and porphyrins. In addition, calculated properties for an additional 27 molecules are used as an independent test on the quality of the parametrization obtained. Reference data are taken from eitherexperiment or previous ab initio calculations or evaluated using ab initio or density functional theory. For comparison, analogous results for all 59 molecules are obtained using the semiempirical PM3 and MNDO/d methods. Both AM1 and MNDO/d are found to be robust and widely applicable for magnesium compounds while the applicability of PM3 appears significantly restricted. MNDO/d appears the method of choice for ionization potentials and heats of formation while, reflecting our parametrization priorities, AM1 appears the method of choice for geometrical properties, especially those of magnesium porphyrins.