The present study compares the performances of the regression models developed by the response surface methodology combined with the full factorial, Box-Behnken or face central composite designs applied for the ultrasound-assisted KOH-catalyzed methanolysis of sunflower oil. While all models led to similar optimal reaction conditions, the models based on the simpler designs had the smaller corrected Akaike information criterion values, the insignificant lack of fit and the more favorable statistical criteria than the model based on the full factorial design. Including fewer experiments, the Box-Behnken design can be recommended for the optimization of ultrasound-assisted biodiesel production processes.