Leaf area index (LAI) is an important biophysical parameter, and is the critical variable in many ecology models, productivity models and carbon circulation study. Based on the field experiment data, an evaluation of soybean LAI retrieval methods was conducted using NDVI (normalized difference vegetation index) and RVI (ratio vegetation index), principle component analysis (PCA) and neural network (NN) methods, and the estimate effects of three methods were compared. The results showed that the three methods have an ideal effect on the LAI estimation. R2 of validated model of vegetation indices, PCA, NN were 0.753 (NDVI), 0.758 (RVI), 0.883, 0.899. PCA and NN methods were better with higher precision, and PCA method was the best, as its RMSE (0.202) was slower than the two vegetation indices (RMSEs of NDVI and RVI were 0.594 and 0.616) and NN (RMSE was 0.413) method. While the LAI was small, vegetation indices were obvious for removing the noise from soil and atmospheric effect and obtained the good evaluation result. PCA showed better effect for all LAI. LAI affected the estimating result of NN method moderately. As for the NN method, modeled LAI value and measured LAI regression formula slope was the nearest to 1 with R2 of 0.949, which showed a great potential for LAI estimating. As a whole, PCA and NN methods were the prior selection for LAI estimation, which should be attributed to the application of hyperspectral information of many bands.