Spectral data was captured for selected different growing and color of nine kind of tea by using ASD Field Spec Hand Held 2 spectrum analyzer. The raw data was pre-processing to decrease noise and improve acquisition. Multiple linear regression analysis, SG and PAC were employed to analysis the data. The optimization model was established to predict the best time of harvesting. The principal component characteristic bands of 520, 765, 821, 940 nm was obtained by smoothing, the first derivative, the second derivative and the pretreatment process. The determination coefficient is close to 0.99 for this model using least square method. By analyzing other 23 kinds of tea to verity the model, the correlation coefficient of real and predictive value is 0.97 as well as the average relative error of 2.99%.