We focus on model-driven analysis of images. How can such models be generated efficiently and how is knowledge about expected image content integrated before and during an analysis? We use deformable models as they are very flexible for representing shape and appearance of structures. Research topics are, how to generate such models from few samples, how to use these models for search and classification of objects, how to represent non-linear variation and how to integrate user knowledge efficiently. Application areas for our models are segmentation, object detection, object classification, image registration and tracking of objects.