Abstract: Based on the demand of precision in injection timing control, BP neural networks structure is established and several groups of experimental data are used as the training sample, an injection timing MAP is concluded. Using the sum calculation error as the measure indexes, the scope of sample number and minimum training-sample number meeting the control demand are discussed by quoting the random error synthetic method. Optimal injection timing test derived from minimum training-sample number is made, and it is proved that the diesel engine performance levels under the optimal injection timing are optimum. Practices prove that this method can meet the precision demand of the control system and release the injection timing experimental quantity.