The effective region was segmented from the hyperspectral image of citrus leaf by threshold method with the average spectrum extracted and used to describe the corresponding leaf. Based on the different spectral pre-processing methods, the prediction models of three photosynthetic pigments (i.e., chlorophyll a, chlorophyll b, and carotenoid) were calibrated by partial least squares (PLS), BP neural network (BPNN) and least square support vector machine (LS-SVM). The LS-SVM model for chlorophyll a was established based on multiplicative scatter correction (MSC), and the correlation coefficient (Rp) and the root mean square error of prediction (RMSEP) were 0.898 3 and 0.140 4, respectively. The LS-SVM model for chlorophyll b with Rp=0.912 3 and RMSEP=0.042 6, was established based on standard normal variable (SNV). The PLS model for carotenoid was established with Rp=0.712 8 and RMSEP=0.062 4 based on moving average smoothing (MAS), but the result was no better than the other two. The results illustrated that these three photosynthetic pigments could be nondestructively and real time estimated by hyperspectral image.