158 samples of comminuted green tea were scanned by the near infrared spectroscopy analyzers. Using orthogonal experimental design, the calibration models of free amino acids content in green tea were gained with the correction methods of PCR, PLS and MPLS under different spectral data preprocessing methods separately. The models were evaluated by the objective function method. The optimal correction methods and spectral data preprocessing methods were determined, and the applicability of the models was validated. The experimental result is that using MPLS could get the optimal model. And its spectral data preprocessing method is SNV+Detrend, 1, 1, 4, 4. Its objective function value is 94.25%; the relative standard deviation of validation is 6.39% (less than 10%). The results showed that using orthogonal experiment design could comprehensively evaluate the different correction methods and spectral data preprocessing methods on the models, and NIRS could achieve quantitative and fast detection of free amino acids content in green tea. 