With the large-scale development of the technology—Gene Expression Profiles, the diagnostic method based on gene expression profiles is now becoming a quick and effective method in clinical medicine. But because of gene expression data’s high dimension, small sample size and large noise, extracting the information about cancer correctly becomes the key point. In this paper, the gene expression data of colon tumor as an example put forward the mixed information gene extraction method combining Fisher Weight Function, discrete Fourier transform and principal component analysis and take multiple Logistic regression analysis together with Bayesian decision as classifier to do tumor classification and detection. The experiment results show that, the accuracy of 96.80% is achieved on CV recognition for colon cancer’s data set using this method.