Mid-infrared spectroscopy was used to measure the concentration of ammonia (NH3) in soil headspace. The result showed that a spectrum with stronger signal and less noise could be got when using the mirror velocity of 0.16 cm/s and the resolution of 4 cm–1. The spectra of NH3 of different concentrations were obtained with the selected parameters, and the absorption peak within 900–1 000 cm–1 was appropriate for quantitative analysis. Linear regression models were established between NH3 concentrations and peak height, peak area and the first principle respectively, and the corresponding decision coefficients (R2) were greater than 0.9, indicating that all of them had significant linear correlation with NH3 concentration. The concentration of NH3 in soil headspace could be predicted quickly using mid-infrared spectroscopy combined with principal component regression.