Visible-near-infrared spectroscopic characteristics of soils sampled in cropland in oil well region were analyzed and these soil samples were meshed by 2 mm, 0.25 mm, 0.15 mm. Reflectance spectra was pretreated by first derivative, continuum removal, and multiple scattering correction before study on relationship between spectral reflectance and soil petroleum hydrocarbon concentration. Partial least square regression models for predicting soil petroleum hydrocarbon concentration were built on the basis of the full wavebands from 350 to 2 500 nm. The results showed that the most sensitive wavebands of RAW, continuum-removal, and multiple-scattering-correction spectra were located in the region from 350 to 600 nm with 0.05-level significance, and that those of first-derivative spectra were located at 2 280 nm (|r| = 0.81, P＜0.01). Estimation accuracy of partial least square regression models for predicting soil petroleum hydrocarbon concentration was increased by spectral pretreatment, ; nevertheless, effects of soil milling on estimation accuracy depended on spectral pretreatment. In this study, 2 mm and first derivative were respectively considered as suitable soil particle size and spectral pretreatment in building models for predicting soil petroleum hydrocarbon concentration on the basis of visible-near-infrared spectroscopy and spectral analysis, which provides a novel alternative of traditional analysis methods.