Pretreatment of near-infrared spectral data is the basis of feature extraction, quantitative and qualitative analysis and establishment of models, it plays a significant role in obtaining the data and get reliable results. The purpose of the paper is compared the advantages and disadvantages of the S-G, derivative and multiple algorithm methods of spectral preprocessing through the example of apple leaves. S-G algorithm can smooth the data relatively better, but we must according to the specific circumstances of the case while chose the width of the window and the order of polynomial; Kernel smoothing is better than S-G in two-ends data processing, but its processing speed is slower than S-G. Derivative algorithm can get more stable reflectance, but it is sensitive to noise, so it need to be used with the smoothing algorithm. Multiple scatter correction can be used effectively to eliminate the translation and offset of baseline. All of above algorithms have been applied in the system of near infrared spectroscopy processing system of leaves and satisfactory result was obtained.