Spectral data fusion， an emerging spectral analysis technology， can effectively combine the detection characteristics of different spectrum technologies. The classification of 288 handkerchief paper samples were studied with data fusion of infrared spectrum fingerprint region and strong signal region of Raman spectrum. A classification model was established on 189 training samples by linear discriminant analysis. Compared with the independent use of infrared spectroscopy or Raman spectroscopy， the accuracy was significantly improved. However， the distinguishing effect was still not ideal. The accuracy of 35 principal components extracted by principal components analysis for linear discriminant analysis can reach 100%. The model accuracy of the 32 principal components extracted from test samples reached 100%， and the classification results were ideal. The spectral fusion technology provides an effective tool for the classification and identification of handkerchief paper in the field of forensic science.