Predictions of distinct gasoline effects on the behavior of an engine performance. Metamodels were generated using radial basis functions and kriging techniques. The “leave-one-out” cross-validation procedure was applied to select metamodels. Most of the prediction residuals were lower than 3%.