The enantioselective ring-opening catalyzed by epoxide hydrolases originating from seven different sources of a series of 2,2-disubstituted oxiranes containing alkyl chains of different lengths, unsaturated (alkenyl, alkinyl) and aromatic groups as well as electronegative heteroatoms at various positions within the side chain was analyzed by quantitative structure-activity relationships. Models for the enantioselectivity were derived with the aid of multiple linear regression analysis (MLR) using several steric and electronic (quantum chemical) descriptors. On the basis of the models derived by MLR nonlinear modeling with artificial neural networks (ANN) was also done. Good predictive performance was observed for both modeling approaches. The models also indicate that different steric and/or electronic features account for the enantioselectivities observed for the individual epoxide hydrolases.