A near-infrared mathematical model for the detection of the total fatty acid (TFA), saturated fatty acid (SFA), and unsaturated fatty acid (UFA) contents in infant formula was established using Fourier near-infrared (NIR) spectroscopy combined with partial least squares (PLS) regression prediction. The reliability of the model was verified by cross-validation and external validation. Different wavelengths and different correction algorithms, including smoothing, vector normalization, first derivative, second derivative, and multiple scatter correction (MSC), were used to process the NIR spectra. The correction model correlation coefficients (R2) for TFA, SFA, and UFA contents were 0.9337, 0.9374, and 0.9020, respectively. The coefficient residual predictive deviations (RPDs) were 3.63, 3.65, and 2.90, respectively. These data demonstrated that this NIR mathematical model had good predictive performance. Twenty collected infant formula samples were predicted using the established model. Paired sample t test analysis showed that the chemically measured and predicted values of TFAs, SFAs, and UFAs had no distinct statistical differences.