Prediction of mechanical power losses of gear pairs requires elastohydrodynamic lubrication (EHL) analyses to be carried out at every discrete contact position. This makes gear efficiency models computationally demanding. As a remedy to this problem, a method is proposed in this study to derive EHL-based friction coefficient and rolling power loss formulae to be used in gear efficiency models. This method employs a large number of EHL analyses covering a full matrix of all key contact parameters, namely normal load, rolling and sliding velocities, radii of curvature, lubricant parameters and surface roughness amplitudes, within typical ranges dictated by gear contacts. Linear regression of the results of the EHL analyses yields friction coefficient and rolling power loss formulae suitable for gear efficiency models. In the EHL model, the hydrodynamic fluid and asperity contact zones are treated in a unified manner. The sliding friction is computed as the sum of the viscous shear within the lubricant and the boundary friction at the local asperity contact spots. The rolling power loss induced by fluid pressure gradient is formulated to include the entire contact zone instead of only the inlet region, since the pressure gradient is substantial within the contact for rough surface condition. The proposed method is demonstrated using an example turbine fluid. Typical measured gear surface roughness profiles are used in the analyses. At the end, the results from the regression formulae are compared to actual EHL predictions to assess their accuracy under various contact conditions.