This study explored the feasibility of near-infrared (NIR) quantitative and qualitative models for soybean inorganic phosphorus (Pi), which is complementary to phytic acid, a component of nutritional and environmental importance. Spectra, consisting of diffuse reflectance (1100-2500 nm) of ground meal and single-bean transmittance (600-1900 nm) of whole seed, were collected on 191 recombinant inbred soybean lines. Partial least-squares regression models were individually developed for soy meal diffuse reflectance, single-bean transmittance, and averaged (24 beans/line) whole seed transmittance data. The best performance was obtained with diffuse reflectance data, in which the standard errors (rmsd) were 263 and 248 mg/kg for cross-validation and validation sets, respectively. Model accuracy was lower for the 24-bean average transmittance spectra and still lower for single beans. Despite the overall poorer modeling ability of Pi with respect to the common macronutrient NIR regressions, such as those for protein and oil, this technique holds promise for use in breeding programs.