Early disease detection is a major challenge in horticulture. Integrated Pest Management (IPM) (Stern et al., 1959;Altner et al., 1977) combines prophylactic, biological and physical methods to fight bioagressors of crops while minimizing the use of pesticides. This approach is particularly promising in the context of ornamental crops in greenhouses because of the high level of control needed in such agrosystems. However, IPM requires frequent and precise observations of plants (mainly leaves) which are not compatible with production constraints.Our goal is early detection of bioagressors. In this paper, we present a strategy based on advances in automatic interpretation of images applied to leaves of roses scanned in situ. We propose an cognitive vision system that combines image processing, learning and knowledge-based techniques. This system is illustrated with automatic detection and counting of a white fly (Trialeurode Vaporariorum Westwood) at mature stage. We have compared our approach with manual methods and our results showed that automatic processing is reliable. Special attention was paid to low infestation cases which are crucial to agronomic decisions.