Near-infrared (NIR) and mid-infrared (MIR) diffuse reflection spectra were compared and evaluated for hydrocarbon potential generation of source rocks.Near-infrared diffuse reflectance often exhibits significant differences in the spectra due to the non-homogeneous distribution of the particles, so the signal-to-noise ratio of NIR is much lower than MIR.It is too difficult to get accurate results by NIR without using a strong spectral preprocessing method to remove systematic noise such as base-line variation and multiplicative scatter effects.In the present paper, orthogonal signal correction (OSC) and an improved algorithm of it, i.e.direct orthogonal signal correction (DOSC), are used as different methods to preprocess both the NIR and MIR spectra of the hydrocarbon source rocks.Another algorithm, wavelet multi-scale direct orthogonal signal correction (WMDOSC), which is a combination of discrete wavelet transform (DWT) and DOSC, is also used as a preprocessing method.Then, the calibration model of hydrocarbon source rocks before and after pretreatment was established by interval partial least square (iPLS).The experimental results show that WMDOSC is more successfully applied to preprocess the NIR spectra data of the hydrocarbon source rocks than other two algorithms, and NIR performed as good as MIR in the analysis of hydrocarbon potential generation of source rocks with WMDOSC-iPLS pretreatment calibration model.