Abstract: In this paper, diffuse reflectance spectroscopy was applied to detect and research the pesticide concentration on the surface of leaves. The optimal hands for 350～1 900 nm were obtained, the normalization of standard deviation, the average filtering of three slides and the first derivative combination were selected as combination pretreatment method, five kinds of mathematical models with the applications of stepwise regression analysis, principal component, principal components combined with the artificial neural network, partial least squares, partial least squares combined with the artificial neural network were established. The results indicate that the root mean square error of cross-validation of the prediction of five algorithms were 0.067, 0.061, 0.059, 0.039, 0.056 respectively. The partial least squares method has high prediction precision. Considering the effects of different crop types to the leaves prediction precision, three kinds of plant leaves for object:fatsia japonica, brassica napus and green vegetables were selected by using the partial least squares method, the correlation coefficient between the prediction values and the truth values in the prediction set were 0.994, 0.974, 0.929 and the root mean square error of cross-validation of the prediction set were 0.039, 0.050, 0.075. The results indicate that different kinds of crop leaves have less influence on the measurement of solution concentration and it is feasible to test the pesticide concentration with diffuse spectroscopy technology.