This research collected the transmission hyper-spectral data of eggs with hyper-spectral imager. Haugh unit value was used as freshness norm. With the help of MATLAB and SAS software combined with stechiometry method, the hyper-spectral data of sample eggs was analyzed and processed. The prediction model of egg freshness was established based on hyper-spectral technology. The research chose the band range from 500 to 1 000 nm as sensitive band. The hyper-spectral data of abnormal samples were removed by using mahalanobis distance. Differential correction was done on hyper-spectral data. After the comparison, there was a high linearity between the second-order differential data of hyper-spectra and haugh unit value. Therefore, this paper conducted a further research on the second-order differential data of hyper-spectra. And it was treated with wavelet denoising, smoothing and standardizing. This paper chose the newly proposed CARS variable selection method to do dimensionality reduction on hyper-spectral data. And thirty-two characteristic parameters were extracted. They were used to establish partial least square prediction model based on all band and multiple regression model based on characteristic parameters on white shell eggs. The correlation coefficients of white shell eggs were 0.88 and 0.93 respectively, and the corresponding mean square errors being 7.565 and 6.44. Inspections were conducted on PLS prediction model based on all band hyper-spectral second-order differential and multiple regression model based on characteristic parameters by using eggs of validation set. The accuracy rates of these two models to discriminate white shell eggs’ freshness and non-freshness were 100% and 88% respectively.