Light changing and targets overlapped each other were main reasons of affecting fruit recognition accuracy under natural condition. In order to reduce both effects, a kind of fruit recognition algorithm based on multi source images fusion was studied. On the basis of image registration, H component image and amplitude image were selected as source images for fusion. The use of fuzzy logic in pixel level fusion was related to the weighted averaging approach where the weights were determined by using the fuzzy inference system (membership function and fuzzy rules). According to the law of fruit area distribution in the fused image, an head threshold detection algorithm based on histogram was presented，so as to get the best fruit segmentation results. According to statisitical properties of range image, a solution for overlapped fruits recognition using layer segmentation algorithm was designed. The experimental results showed that information on multi-source image fusion was used for fruit recognition and location more accurately and robustly than single image, overlapped fruit recognition rate was from 83.67% to 94.22%.