Methanol content and viscosity of methanol diesel oil were quantitatively detected by Raman spectroscopy. As the detected objects, 93 methanol diesel samples were divided into the calibration set (72 samples) and the prediction set (21 samples). The spectral full cross validation partial least squares (PLS) model were analyzed and compared with different pretreatment methods; The spectral data obtained by the optimal pretreatment method is used as inputs, combining with the successive projections algorithm（SPA）to establish different regressive correction models, and these models were comparatively analyzed. The results show that the multiplicative scatter correction partial least squares (MSC-PLS) model for prediction of methanol content is optimal. In this model, the calibration correlation coefficient RC is 0.9761, the interactive validation correlation coefficient RCV is 0.9551, the root mean square error of calibration set (RMSEC) is 1.5089, and the root mean square error of cross-validation（RMSECV）is 2.0630. For viscosity determination, MSC-PLS model also has the best prediction performance, whose RC is 0.9794, RCV is 0.9580, RMSEC is 0.0907 mPa·s, and RMSECV is 0.1292 mPa·s.