Fast, accurate and non-destructive identification of seed viability is a critical factor in the safe preservation of germplasm resources. In this paper, we have used yellow soybeans of different preservation period as samples, and established a mathematical model of non-destructive determination for its viability by employing Fourier transform near-infrared spectroscopy (FT-NIRS) in combination with partial least squares (PLS). Simultaneously, we have compared and analyzed the prediction performance of the model by using different spectral-pretreatment methods and different band ranges. Results indicate that the model has higher prediction accuracy in 4000~10000nm full-band range. After preprocessing these spectra data with Savitzky Golay second derivative and standardized pretreatment, we’ve found that the PLS model is the best for viability non-destructive determination. The correlation coefficient of calibration set samples is 0.937, and the correlation coefficient of prediction set samples is 0.902. RMSEC and RMSEP are 2.190 and 2.684 respectively. Therefore, the prediction accuracy of the model is close to that of normal germination method, and can meet the requirement in rapid non-destructive identification of seed viability, which provides theoretical basis for rapid non-destructive determination of seed vigor in the future.