While sliding at seismic slip-rates of ∼1 m/s, natural faults undergo an abrupt decrease of shear stress called dynamic weakening. Asperity-scale processes related to flash heating and weakening and, meso-scale (mm-cm) processes involving shear across the bulk slip-zone, related to frictional melting or viscous flow of minerals, have been invoked to explain pronounced velocity-dependent weakening. Here we present a compilation of ∼100 experiments performed with two rotary shear apparatuses. Cohesive rock cylinders of basalt, gabbro, granitoid rocks and calcitic marble were sheared at various values of effective normal stress (σneff = 5–40 MPa), target slip-rate (Vt = 0.1–6.5 m/s) and fluid pressure (Pf = 0–15 MPa). To account for the uncertainties of constitutive parameters, we introduce a norm-based optimization procedure on a set of model parameters by comparing the shear stress evolution inferred from the proposed weakening models with the shear stress measured during the experiments. We analyze the fit to experimental data of each weakening model and we discuss a composite model in which two weakening mechanisms (namely flash heating and bulk melting, flash heating and dislocation/diffusion creep) are used to test the hypothesis that they match the shear stress evolution in different slip ranges. We found that for slip smaller than a slip-switch distance δ0, the weakening is better described by mechanisms occurring at the asperity scale whereas for larger slip values the bulk model performs better. The inferred δ0 values decrease with normal stress suggesting that during earthquakes bulk mechanisms can govern shear stress evolution after a few centimeters of slip.