To explore the potential of thermal infrared hyperspecra for retrieving sand content in soil, the sandy soil was measured using a 102F Fourier Transform Infrared Spectroradiometer (FTIR), and the characteristics of sandy soil’s emissivity spectra were discussed based on correlation analysis and principal component analysis. Moreover, the sand contents were predicted using two modeling methods: Partial least squares regression (PLSR) and principal component regression (PCR). The results show that the Reststrahlen feature(RF) of SiO2 is obvious in the emissivity spectra of sandy soil with two large asymmetrical absorption troughs near 8.13 and 9.17 　μm and two small troughs in the region of 12～13　μm. Soil emissivity becomes lower when sand content increases, this trend is more evident especially in the regions of 8～9.5 　μm　and 9.5～10.4　μm of which correlation coefficients are above 0.65 and 0.5 respectively, and these two regions can account for 84.07% of total emissivity variance. Predictive precision varies significantly when sand content is predicted by different modeling methods or spectral variables. The PLSR model can achieve the highest predictive precision by using first-order derivative spectra, and it’s RMSE of modeling and prediction is 0.45 and 0.53 respectively, and the R2, 0.990　7 and 0.983　6, which means that the thermal hyperspectra has promising potential for retrieving sand content in soil.