Near-infrared spectroscopy (NIRS) can detect soil total phosphorus in agricultural environments. Considering the soil moisture and particle size on total phosphorus prediction, we applied NIRS to the detection of soil samples with different soil moistures and particle sizes. Thus the effect of soil moisture and particle size was analyzed quantitatively and qualitatively. The procedures to remove the effect of soil moisture and particle size on total phosphorus prediction were also described. First, the near-infrared reflectance spectra of soil samples with different soil moistures and particle sizes were obtained and the absorbance values were determined. Next, the original spectra were corrected by using moisture absorbance index (MAI) and hybrid correction to counteract the effects of soil moisture and particle size, respectively. Absorbance of soil samples showed high correlation with soil moisture at wavelengths of 1,450 nm and 1,940 nm. MAI is a tool for normalizing the original spectral data so as to correct for soil moisture. Hybrid correction is based on the superposition of NIR spectra, and a particle size different from that of the original soil samples is generated. This is an effective means of correcting for the effect of soil particle size. Finally, using the corrected absorbance values at eight wavelengths (655, 722, 1,055, 1,255, 1,467, 1,678, 1,890, and 2,246 nm), the soil total phosphorus prediction model was built based on LS-SVM. Compared with the model used for original spectral data, the new model exhibited higher accuracy and stability. Results showed that MAI and hybrid correction are effective for correcting for soil moisture and soil particle size during the prediction of soil total phosphorus.