Abstract: The HSI color model was used to describe the color feature of the apple in the paper. Four chroma average values were used to substitute the apple chroma from the apple color histogram. The apple grading system based on BP neural network was established. In the system, there were 4 nodes of chroma in the input layer, 2 nodes of apple types in the output layer and 5 nodes in the implicit layer. The experimental results showed that 95% apples could be graded precisely and the system could meet practical demands.