In order to acheieve non-destructive detection and gading of egg freshness, an experimental system was set up based on computer vision. Then egg internal substance transmission images were acquired. After pre-processing, H, S, I, a, b values were extracted. The egg shell color information (a*, b*) was also measured. And the weight of egg was measured using electronic balance and the height of egg albumen was measured using height vernier calliper. Egg freshness was calculated according to its weight and albumen height. The linear regression model for egg Hough unit and egg information (H, I, S, a, b, a*, b*, a-a*, b-b*) was established by SAS. Afterwards the 3 parameters (H, I, b) which is greatly correlated with egg freshness (HU, Egg Hough unit) was reserved. With 3 parameters (H, I, b) of input, the best BP neural network model (3 input nodes, 15 hidden nodes, 4 output nodes) was established by using Matlab. On the BP neural network model of detecting the egg freshness, the automatic detection system was designed, which can immediately show the results according to the egg′s color data after the network initialization. The results showed that the grading accuracy by using computer vision and BP neural network for egg freshness exceeds 90%.