Free iron oxide content is an index of soil weathering degree, soil-forming condition and environment, soil classification and soil age estimation. This paper investigated 91 soil profiles containing 398 samples in XuanchengPrefecture of Anhui Province, the spectral reflectance of soil samples was measured over the range 350 – 2500 nm, the free iron oxide content was analyzed by chemical means. In order to determine the relationship between the free iron oxide and spectral data, the spectral data were transferred to first derivatives of the reflectance (FDR) and absorbance (Log (1/R)) other than reflectance (R). This paper compared the performance of two calibration methods, namely partial least squares regression (PLSR) and back propagation neural network (BPNN) analyses for the accuracy of measurement of three spectral dataset, namely R, FDR and Log(1/R), the spectral response of the free iron oxide also was determined. The results showed that traditional modeling methods could not accurately predict free iron content (R2<0.6, RPD<1.5) when its contents were lower than 20 g/kg. Compared to the R and log (1/R), the FDR which was implemented in the models indicated a poor predictive ability.