A quantitative structure-activity relationship ( QSAR) model is presented in which both the independent and dependent ( response) variables are derived from density functional theory (DFT) calculations on a large set of 14-electron complexes, LCl2Ru=CH2, with different dative ligands, L. The multivariate model thus correlates the properties of the 14-electron complexes with a calculated measure of activity, with modest computational cost, and reproduces the experimental order of activity for the Grubbs ruthenium catalysts for olefin metathesis. The accuracy and applicability of the model is to a large extent due to the use of highly specific geometric and electronic molecular descriptors which establish a direct connection between activity and chemically meaningful donor ligand properties. The ligands that most efficiently promote catalytic activity are those that stabilize the high-oxidation state (+4) metallacyclobutane intermediate relative to the ruthenium-carbene structures dominating the rest of the reaction pathway. Stabilization of the intermediate is ensured, among others, through ligand-to-metal sigma donation, whereas metal-to-ligand pi back-donation destabilizes the intermediate and lowers catalytic activity. A bulky dative ligand drives the reaction toward the less sterically congested metallacyclobutane species and thus contributes to catalytic activity. The multivariate model and the high-level descriptors furthermore provide practical handles for catalyst development as exemplified by the suggestion of several new donor ligands predicted to give more active and functional group tolerant ruthenium-based catalysts. The present strategy holds great promise for broader screenings of olefin metathesis catalysts as well as for development of homogeneous transition metal catalysts in general.