Near-infrared spectroscopy (NIRS) was used for quantitative determination of free amino acid contents in tea samples. Two spectral preprocessing methods including continuous wavelet derivative (CWD) and standard normal variate (SNV) were used for spectral transform. Partial least squares regression (PLSR) was used for modeling. Monte Carlo uninformation variable elimination (MCUVE) and successive projections algorithm (SPA) were used for optimizing the modeling variables. It was shown that CWD-SNV method could effectively improve spectral quality, and eliminate translation error. MCUVE-SPA method could greatly improve the precision of model, and compress the modeling variables. The correlation coefficient of prediction ( Rp ) and root mean square error of prediction (RMSEP) of analytical models were optimized from 0.851 and 0.117 to 0.895 and 0.107，and modeling variables were reduced from 4148 to 18. NIRS combined with chemometrics could get a better analytical model when amino acid contents exceeded 0.1%. It can provide a fast and simple analytical procedure for the determination of low contents amino acid.