An optimization of the thermal behavior of a high-power salient-pole electrical machine is presented. Temperatures are calculated with the lumped method, which provides the thermal trends with relatively low computational cost. This model is used to define an aggregated objective function of our nonlinear thermal optimization problem by combining the mean solid temperature with the maximum temperature criteria. The 13 design variables correspond to the main volumetric flow rates in the electrical machine, which are bounded and subjected to a nonlinear constraint, assuming a fixed geometry. Two MATLAB optimization algorithms were tested: the Active-set (FMINCON solver) and the genetic algorithm (GA). Due to the strong nonlinearities of the model and the resulting nonconvex optimization problem, the GA is likely to give better results. Minimizing the mean solid temperature was demonstrated to be more important than the maximum temperature criterion. A strategic flow configuration is found to send fresh air to the second half of the cooling circuit, where air usually arrives heated. This optimal configuration provides better cooling than its current modeled configuration. This methodology should be of interest during the development phase.