Abstract: Tea polyphenol, a vital indicator used for the detection of tea quality, is of great significance to quickly detect the tea polyphenol content via near infrared spectroscopy. In this paper, the near infrared spectroscopy in combination with partial least squares (PLS) was adopted to establish the rapid analysis models by near infrared for tea polyphenol content of powdered and complete tea samples respectively, using 144 black tea samples as the study objects, revealing that the prediction model established by SNV+first derivative+Savitzky-Golay smoothing combined with PLS had the optimal effect in the results. The correlation coefficient (r) was 0.9990 and the root mean square error of calibration (RMSEC) was 0.165% of the training set, while the r was 0.9243 and the root mean square error of prediction (RMSEP) was 0.972% of the prediction set in powered tea samples. At the same time, the r was 0.9967 and the RMSEC was 0.310% of the training set, while the r was 0.9541 and the RMSEP was 0.870% of the prediction set in complete tea samples. The results showed that the PLS model for complete tea samples was better than that for powdered tea. Therefore, rapid and nondestructive detection of tea polyphenols in black tea can be achieved by near infrared spectroscopy.