Due to the advantages of high detection speed, real-time measurement, and non-destructive process, spectroscopy analysis technique is widely used in every field and has becomes one of the most effective methods in modern agricultural control. In this paper, we collected soil samples with different nutrient contents, measured the infrared attenuated total reflection (ATR) and diffuse reflection infrared Fourier transform (DRIFT) data of every sample in order to built models and predicted the content of total carbon, total nitrogen and available nitrogen quantitively in soil by using pretreatment and partial least squares(PLS) methods. It indicated that the determination coefficients of PLS models improved after using standard normal variate transformation (SNV). For the ATR and DRIFT models, the determination coefficients of total carbon were improved to 0.826, 0.919 and 0.841, 0.928 of total nitrogen respectively. By combining with ATR and DRIFT spectra, the prediction accuracy became higher than that of using ATR or DRIFT spectra. The coefficient of determinations of the total carbon, total nitrogen and available nitrogen models were 0.942, 0.951 and 0.919 respectively. It provided a reasonable method to improve the analytical ability in mid infrared spectroscopy.