Research on object recognition has focused mainly on bottom-up processes that analyze the visual input; the potential role of top-down processes has been relatively neglected. We propose a framework that views object recognition as discrimination between probable alternatives – an iterative process in which bottom-up and top-down processes interact, and in which attention plays a critical role. Although elements of this framework have been discussed by others, to our knowledge this is the first attempt to test these ideas empirically. A series of experiments tested the basic hypothesis derived from this framework: In the course of object recognition attention is directed to distinguishing features – features that are the most informative and diagnostic of object identity in a particular context. Observers discriminated between artificial fish that differed in one distinguishing feature, such as mouth or tail-fin (in separate conditions). Differences in the allocation of attention to distinguishing and non-distinguishing features were examined using primed-matching tasks (Experiments 1, 2), in which identity matching of two target figures is faster if similar stimuli have previously been attended than if they have not, and using visual-probe and spatial-cueing tasks (Experiment 3, 4), in which visual processing at a specific location is enhanced when attention is allocated to that location. The results showed that (1) both global and local distinguishing features yielded a greater amount of priming than non-distinguishing features; (2) detection of a visual probe was more accurate when the probe appeared near the location of a distinguishing feature than near the location of a non-distinguishing feature; (3) object (fish) recognition was faster when attention was pre-allocated to the location of a distinguishing feature than to the location of a non-distinguishing feature. These results support the hypothesis that attention is allocated to distinguishing features during object recognition.