Aimed at determining the maturity degree of citrus fruit non-destructively, three tone images were generated by dividing hue range 30°~120° of fruit image equally to three parts. Scale invariance and spectra of multi-fractal were analyzed. Height and width of multi-fractal spectra were extracted as features of color and luster of pericarp of fruits, and were set as input of BP neural network. With total soluble solid contents as the output of network, the neural network maturity degree model mapped fruit image into degree of maturity. The correctness of inspection test was 82%, which showed that maturity degree of citrus fruit could be detected non-destructively based on multi-fractal spectra. 