Near-infrared spectroscopy was used to qualitatively discriminate the adulteration in whole egg powder and to analyze the concentration of adulteration quantitatively. Six different pre-processing methods were used to pre-process the raw spectral data. The competitive adaptive reweighted sampling （CARS）， successive projections algorithm （SPA） and combined CARS-SPA algorithms were used to filter the spectral data for characteristic wavelengths. Ensemble learning （EL） model was established to qualitatively discriminate the egg powder adulterated. The partial least squares regression （PLSR） model was established to quantitatively analyze the concentration of adulteration. The results showed that first order derivative was the best pre-processing method in qualitative discrimination. The CARS algorithm had the best effect of feature filtering. The EL model had an overall discrimination accuracy of 98.18% for adulterated samples， with the discrimination accuracy of differently adulterated samples of more than 97.78%.MSC was the best pretreatment method in quantitative analysis. CARS algorithm had the best effect of feature filtering. The prediction set correlation coefficient of PLSR adulteration concentration prediction models for one-component adulteration， two-component adulteration， three-component adulteration， and all adulterated samples adulteration was 0.958 5， 0.931 2， 0.945 6， and 0.955 8， respectively， with the root mean square error RMSEP of 4.689 1， 5.813 4， 4.604 1， and 3.802 9， respectively. It is indicated that the near-infrared spectroscopy can be used for detecting the adulteration in whole egg powder. It will provide a reference for regulatory agencies to detect adulteration in whole egg powder.