Objective To establish a rapid detection model for the nutritional components of Oncorhynchus mykiss based on near-infrared spectroscopy (NIR). Methods Firstly, near-infrared spectral data of 200 Oncorhynchus mykiss were collected. Then, the content of 3 kinds of nutritional components (moisture, fat and protein) of each Oncorhynchus mykiss were determined by national standard methods. The near-infrared spectral data and nutritional component data were matched one by one. The near-infrared rapid detection model was established by combining NIR with partial least squares (PLS), and the best detection model was screened out. Results The pretreatment method of the moisture content detection model was multiple scattering correction (MSC), and the optimal model was obtained when the wavelength range was 4000-10000 cm^-1. The pretreatment method of the fat content detection model was standard normal variable (SNV), and the optimal model was obtained when the wavelength range was 5000-7144 cm^-1 and 7404-10000 cm^-1. The pretreatment method of the protein detection model was second derivative (ds2)+SNV+Savitzky-Golay smoothing (sg9), and the optimal model was obtained when the wavelength range was 4100-5100 cm^-1 and 5400-9000 cm^-1. The Q value, the correlation coefficient of the correlation coefficients, and the correlation coefficient of prediction of the optimal model were all relatively large, and the standard deviation of square error corrected and standard error of cross validation were close to each other, which met the best modeling principle. Conclusion The best model is verified by the prediction set that do not participate in the modeling. The absolute deviation between the predicted value and the true value (determined by the national standard method) is no more than 5.7%. This indicates that the model can be used for the detection of three nutritional components of Oncorhynchus mykiss, and can achieve non-destructive and rapid detection of the nutritional components of Oncorhynchus mykiss, saving detection costs and shortening the detection cycle.