Hyperspectral images of six varieties of Kentucky bluegrass were acquired using hyperspectral imager (550-1 000 nm) and the leaf spectral properties were extracted. Wilks’ lambda stepwise method was used and 9 optimal wavelengths were selected from the original 94 wavelengths and the discriminant models for varieties identification of Kentucky bluegrass were built based on Fisher’s linear discriminant function. The results showed that the Fisher’s linear discriminant model with 9 wavelengths achieved classification accuracies of 100% for both training and testing samples. While for the models with three wavelengths and six wavelengths, classification accuracies reached 83.3% and 96.7% for the testing samples, respectively. It indicates that hyperspectral images combined with discriminant analysis might be a good method to identify the varieties of Kentucky bluegrass.