Four varieties of tea were collected from different areas in China including jasmine tea, Kuding tea, Longjing tea and Tieguanyin. A total of 120 samples (30 samples for each variety) were prepared. The original samples spectra were obtained using NIRSystem6500 analyzer. Tea was analyzed qualitatively and quantitatively by near infrared spectroscopy technology. Principal component analysis and discriminant analysis were used to distinguish the four varieties of tea. The optimal calibration model for qualitative discrimination was established according to comparison of different spectral data pretreatment methods and the uncertain factor coefficients. Quantitative analysis models for moisture content, tea polyphenol and caffeine in tea were developed with modified partial least square. The results show that the accurate recognition rate for the four varieties of tea in the validation set reached 100%. The coefficients of determination (Ｒ２ｐ) and relative prediction deviation (RPD) of independent validation sets were more than 0.91 and 3.0, respectively. It is concluded that NIRS can be used as a rapid method to detect the variety and chemical components in tea.