The Loss-Surface (LS) model, commonly used for the prediction of iron losses in electrical machines, is reformulated for better accuracy and reliability when it is extrapolated beyond the parametric identification region. First, the dynamic hysteresis behaviors of typical soft magnetic materials are analyzed by both measurements and FE simulations. They are then described by the LS model using different mathematical formulations. The most relevant one, which requires 14 parameters to reproduce and extrapolate any loss surface, will be presented in this paper. The newly reformulated LS model is tested, in the last part, to estimate the iron losses of intrinsic measurements as well as of an electrical machine. In the first case, the results obtained of four different materials are generally within 5% error. And in the latter case of a high speed permanent magnet synchronous motor, by performing a study on the effects of manufacturing methods (punching and assemble) with respect to the iron losses, it is shown that the LS model provides an estimation with around 10% of deviation.