Reflectance spectra of winter wheat leaves specimens was acquired with portable spectroradiometer and integral sphere, after pretreatment with the method of multiplicative scatter correction(MSC), the principal components calculated were used as the inputs of artificial neural networks to build the Back–Propagation artificial neural networks model(BP-ANN), which can be used to predict moisture content of winter wheat leaves very well. In the article we made a study of quantitative analysis for moisture content of winter wheat leaves in booting and milk stage. The correlation coefficient(r) of predicted set in booting stage was 0.918, the standard deviation(SD) was 0.995 and the relative standard deviation(RSD) was 1.35%. And in milk stage r= 0.922, SD = 2.24, RSD = 3.37%. The model can truly predict the content of water in winter wheat leaves. Compared with the classical method, the artificial neural networks can build much better predicted model.