A successful detection and classification system must have two properties: it should be general enough to compensate for intra-class variability and it should be specific enough to reject false positives. We describe a method to learn class-specific feature detectors that are robust to intra-class variability. These feature detectors enable a representation that can be used to drive a subsequent process for verification. Instances of object classes are detected by a module that verifies the spatial relations of the detected features. We extend the verification algorithm in order to make it invariant to changes in scale. Because the method employs scale invariant feature detectors, objects can be detected and classified independently of the scale of observation. Our method has low computational complexity and can easily be trained for robust detection and classification of different object classes.