Early detection of cancer with mid-IR spectroscopy is a focal topic in the biomedical engineering field. A method was investigated for the detection of malignant tissues with Fourier Transform Infrared spectroscopy (FTIR) and chemometrics method of pattern recognition. Spectral pretreatment and wavelength range selection were studied to improve the classification effect of soft independent modeling of class analogy (SIMCA). To distinguish the malignant tissues from normal ones, sixty-three stomach tissue samples were investigated in this paper. Forty samples were chosen randomly as calibration set and the rest twenty-three samples as predicting set. The correctness of classification was 91%, which was satisfactory. This approach was proved to be a reliable and practicable method for cancer diagnosis. Benign and malignant tissues can be identified by the combination of FTIR spectroscopy and chemometrics, which is able to be developed as a rapid diagnosis method of cancer.