Green tea is the most popular type of tea in China. The differences of green tea leaves from different categories are very small, and it is hard to distinguish them for non-experts by appearances. Traditional chemical methods are complicated in operation and are destructive to samples and it is difficult to achieve fast and nondestructive analysis. Near infrared spectroscopy (NIR) is a new technology, which is simple, fast, non-destructive, good in reproducibility and can be used for on-line analysis. The differences in the composition and content of the organic components in tea samples would be formed due to different growing environments and panting patterns, which can be measured by the NIR spectra. With the help of NIR spectra, the characteristic information of hydrogen groups can be obtained. The difference information of green tea leaves from different categories can be obtained, and the identification of green tea samples can be achieved. In this study, NIR was applied for nondestructive analysis of green tea leaves from different categories with the aid of chemometric methods. The dataset consists of eight brands of green tea samples. A relation has been established between the spectra and the tea varieties. The data was analyzed with principal component analysis. Furthermore, baseline elimination by continuous wavelet transform was used for improving the accuracy of the method. The wavenumber selection based on standard deviation and relative standard deviation was used to further improve the accuracy. The results show that the total variance explained by the first two principal components in principal component analysis was over 90% and they were enough for further analysis. The result of classification analysis using the original data was poor and cannot be used for the real application. The baseline interference can be eliminated with continuous wavelet transform method and the classification results were improved. The wavenumber selection method based on standard deviation and relative standard deviation consists of two steps. At first, the wavenumbers with standard deviation below 0.005 and the average below 0.01 were removed. Then, the wavenumbers that have large value of relative standard deviation were selected as informative ones, because the larger value of the relative standard deviation, the more variation between the samples. It was found that acceptable classification results can be obtained when several or several tens informative wavenumbers are used. It was found that, the main differences between varieties of tea are polyphenols, amides and amino acids. The results show the classification of different brands of green tea samples can be achieved by the proposed method, which provides a new idea for the rapid analysis of tea samples.