Abstract: Neural network was applied to solving multi-variable inverse heat conduction problem,and a valid numerical model was presented.The processing method of the inputting data and outputting data and the determining method of the number of inner layers were discussed.By the numerical simulation,the change of the training error and the testing error accompanying the variation of the number of inner layers is developed.The results indicate that the method is practicable and has high precision.In addition,in order to create a neural network which has high level of generalization,test sets for trial and error must be used to monitor the training process.The extension of the method in this paper to other inverse heat conduction problem is straightforward.