Objective To study the method of internal boundary parameters identification of middle ear. Methods The numerical model is created using CT technology. Based on Matlab tools, the neural network for identifying internal boundary is proposed. Results The uniform pressure of 105 dB is applied at the outside of the tympanic membrane, and the harmonic analysis is calculated on the model to take the training samples. The internal condition parameters are identified using the good neural network. Conclusion The investigation shows that the inverse method reveals a fast convergence and a high degree of accuracy.