Inhibitory effect to endocannabinoid system-related human monoglyceride lipase (MGL) and selectivity toward fatty acid amid hydrolase of promising maleimide derived inhibitors were investigated by molecular docking and QSAR study. The essential roles of Ala61, Ser132 and His279 related hydrogen bonds and Tyr204 involved π–π interaction, were emphasized by the docking analysis, which were in good agreement with the experimental observations by far. By performing our new developed self-adaptive genetic algorithm (GA) and artificial neural network (ANN) combined method, as well as multiple linear regression and least squares support vector machine based GA method, significant descriptors were selected to build linear and non-linear models. Strong internal and external validations proved the robustness and effectiveness of docking conformation derived models and that importing descriptors from unrealistic conformations based on geometry optimization is not always appropriate for non-linear modeling. Besides, good linear relation between predicted activities and experimental ones towards rat MGL implicates human MGL and rat MGL may share similar inhibitory mechanism.