Objective The method of rapid detection of main ingredients in infant formula milk powder was studied by near infrared spectroscopy. Methods A total of 100 samples of 12 brands of infant formula were collected, and 10 000-4 000 cm-1 band of near-infrared spectra were collected. The calibration model was established by principal component regression and partial least square method. Results The model predicted the best result by partial least-squares method and standard normal variable transformation pretreatment. The predicted variances of main ingredients, protein, fat, dietary fiber, moisture, ash, carbohydrate and energy were 98.44%, 97.40%, 96.18%, 96.74%, 96.97%, 96.55% and 95.35%, and the estimated standard error were 0.354 2,0.473 8,0.201 4,0.105 8,0.093 61,0.520 7 and 13.64. For external validation of the model, the predictive value of 7 main ingredients of the predicted sample was compared with that of the laboratory test result. Comparison result of relative error and relative deviation were below 5.00%. The result of seven main ingredients met the precision requirements of national standards. Conclusion The method could be used to determine the quantity fraction and quantity of protein, fat, dietary fiber, moisture, ash, carbohydrate and energy in infant formula and the quality.