In order to solve the problems arising from the manual calibration method in the developing process of vehicle automatic transmission control unit (TCU), known as time-consuming, heavy workload, high cost and over-dependence on subjective experience, this article researches on a virtual calibration method based on an approximate model to obtain optimal parameters for TCU. The neural network approximate model is established from the test data chosen with the method of DoE (Design of Experiment). The virtual calibration method is then conducted through Optimal Latin Hypercube Design (OLHD) and multi-island genetic algorithm (MIGA) to search the optimal parameters. By comparing the new calibration method with original manual one on the condition of gear 1 up to gear 2, the result shows that the new method can increase the efficiency significantly.