In order to estimate apple yield in orchard automatically, a yield estimation method was presented which combined image processing and back propagation neural network (BPNN) based on the information of leaves and apples in the tree. Firstly, digital images of apple trees were acquired, including half ripe apples (the apple just turned red) and ripe apples (the apple totally turned red). The actual yield of each tree was weighted in harvest time. Secondly, the fruits and leaves on the image of apple tree were identified. Some useful parameters were extracted from data which were used as input variables, and the actual yield was set as output variable. Finally, BPNN estimation yield model was built and the fitting degrees of this model were 0.9287 and 0.9804 for the half ripe apples and ripe apples, respectively. When this model was applied on samples for yield estimation, the correlation coefficient between model and actual was 0.8766 in the half ripe ones and 0.9606 in the ripe ones. The results indicated that both the two models had good reliability and generalization performance. It concluded that the method presented has substantial potential for apple yield estimation.