Heterogeneous catalysts play a key role in the chemical and energy industries. Despite significant progress in theoretical, computational, and experimental studies, identifying the active sites of alloy nanocatalysts remains a major challenge. Atomic electron tomography (AET) has here been advanced to determine the 3D atomic structure of PtNi and Mo-doped PtNi nanocatalysts for the electrochemical oxygen reduction reaction (ORR) and the experimental structures used as input to first-principles trained machine learning to identify the active sites of the nanocatalysts. By analyzing the structure-activity relationships, an equation (termed the local environment descriptor) was formulated to balance the strain and ligand effects and gain p hysical and chemical insights into the ORR active sites. The ability to determine the 3D atomic structure and chemical composition of realistic nanoparticles, coupled with machine learning, could transform the fundamental understanding of the catalytic active sites and provide guidance for the rational design of optimal nanocatalysts.