Middle infrared spectroscopy combined with chemometrics was investigated for the fast determination of protein of mushroom. 140 samples (35 for each variety) were selected randomly for the calibration set, whereas, 40 samples for the validation set. After some spectrum preprocessing, linear modeling method (PLS) and nonlinear modeling LS-SVM were constructed. Different latent variables were used as inputs of LS-SVM. The optimal models were obtained with 8 LVs based on LS-SVM. The correlation coefficient,,root mean square error of prediction for the best prediction by LV-LS-SVM were 0.9275, 0.25961. The result indicated that middle infrared spectroscopy combined with LV-LS-SVM could be applied as a high precision and fast way for determination protein of mushroom.