A compound method is presented for protein secondary structure prediction. The residue sequence was first substituted with corresponding hydrophobic values. Then the sequential hydrophobic values was processed by using discrete wavelet transform (DWT) for denoising and by using continuous wavelet transform (CWT) for roughly analyzing protein secondary structure. Finally, improved Chou-Fasman method and amino acids propensities were used to predict accurate positions of protein secondary structures. The results showed that the accuracy was better than traditional methods. The computational implementation of the entire method was also included in this paper with data test.