The fatty acid concentrations in maize grain were analyzed with a set of 294 samples including normal inbred lines, high-oil inbred lines and high-oil recombinant inbred lines (RIL). The method of partial least squares (PLS) regression with internal cross validation was employed to develop the measuring models of near-infrared reflectance spectroscopy (NIRS) for concentrations of four major fatty acids, palmitic, stearic, oleic and linoleic acids, as well as oil concentration in maize grain. The NIRS models were accurate for oleic acid, linoleic acid and oil concentrations. The determination coefficients of these models in cross validation were 0.89, 0.88 and 0.91, respectively; the determination coefficients in external validation were 0.86, 0.84 and 0.92, respectively; and the ratio of standard deviation (SD) to root mean square error of validation (RMSEV) in both calibration and external validation sets (RSC(P)) was higher than 2.5. But the models for palmitic and stearic acid concentrations were not accurate enough with determination coefficients in cross validation and external validation lower than 0.80, and RSC(P) lower than 2.5. Further practical validation showed that the predicted results by using NIRS models for oleic acid, linoleic acid and oil concentrations were accurate and reliable, which will be a useful approach to the measurement of a large number of breeding samples during genetic improvement of oil quality and quantity in maize.