The present study proposes a new approach to producing accurate estimates of fall dormancy (FD) in alfalfa in a rapid manner. Using near infrared spectroscopy, the approach produces results fast without causing damage to samples. Near infrared reflectance spectroscopy was applied to measuring the spectra of samples. Then principal component analysis (PCA) was conducted on the measurements. The top ten principal components were selected based on their cumulative contribution rates to build a support vector machine (SVM) model. Detailed analysis and discussions were conducted over their parameter and kernel classifications. The experiment found that when c=0.339 2 and g=32, the accuracy of the predictions of the test set can reach 98.182%. Therefore the approach can estimate the FD in alfalfa in a rapid and accurate manner. Moreover, it was compared with other approaches such as principal component regression, partial least squares regression, BP neural networks, and LVQ neural networks. The comparisons have shown that the PCA-SVM model can effectively address the small-sample-size problem and avoid local minimum.