Non-destructive detection methods of sugar content and valid acidity of Gongchuan Wenzhou citrus fruits were investigated based on computer vision. Factors which influenced the accuracy of detection were studied. Citrus fruit images from computer vision system were cut, the background was removed and conversion from RGB to HSI space was made. These images were segmented according to hue value ranges which are 0°～20°, 20°～40°, 40°～60°, 60°～80°, 80°～100° and 100°～120° hue. Fractal dimensions of each segment were calculated as inputs of BP neural network which modeled sugar content and valid acidity of citrus fruits. Results of 167 test samples showed the correctness for accuracy ±1.5°Brix of sugar content is 66.6175%, for valid acidity, the correctness for accuracy ±0.5 is 73.9275%. From these results it concluded that sugar content and valid acidity of Gongchuan Wenzhou citrus fruits has significant correlation with fractal dimension of hue value of fruit pericarp. Computer vision can be utilized to non-destructively detect these two parameters.