Abstract: In order to improve the prediction accuracy of protein secondary structure，a new network model and its coding method are proposed.Firstly，the structure and connection weights of BP network are evolved simultaneously by using global research ability of GEP.Secondly，the coding method of neural network is improved by integrating the hydrophobic value around the residue.The model is employed to predict 36 nonhomologous protein sequences with 6，122 residues in PDBSelect25，the results show that the proposed model and coding method can efficiently improve the prediction accuracy.