This paper describes a method for sweet pepper picking robots working in greenhouses to distinguish the pepper fruits from the leaves. The fruits of the sweet pepper plant are recognized by image processing techniques, using a parallel stereovision system installed in the robot. However, as fruits and leaves of the sweet pepper have almost the same color, it is very difficult to recognize fruits using only color information. In this paper, we propose a new method using reflections of LED lights. The fruits of the sweet pepper are more reflective than the leaves; therefore, we can identify fruits by assuming that the more reflective parts are the fruits. We perform experiments using the improved image processing algorithm in a greenhouse, and the algorithm indeed improves the recognition ability of the robots.