This work presents an algorithm for estimating the number of apples on trees using images acquired with a standard color CCD camera. The proposed system is capable of correctly identifying and localizing more than 85% of the apples in the images. To achieve this high detection rate, color and texture analyses are combined together with shape analysis. In the first step, pixels with a high probability of belonging to an "apple object" are detected according to their color and texture. In the second step, "seed areas" consisting of connected sets of pixels with a high probability of belonging to an apple object are detected. Each seed area is then extended to cover the entire visible area of the apple to which it belongs. Finally, each blob is segmented into simple components that can either be combined into circles or are discarded, so that each of the resulting circles corresponds to an apple.