The present paper reports two experiments that investigate the critical features of an object shape that automatically elicit recognition. Silhouettes of real objects (targets) and meaningless patterns (fillers) in both canonical and non-canonical formats were presented to subjects, in an attempt to test whether information about the global shape of an object was sufficient for automatic object identification. In Experiment 1, target-filler discriminability was evaluated by means of a reality-decision task. In Experiment 2, subjects had to perform an elongation-decision task, previously shown to be sensitive to the influence of automatically activated object identities (Dell'Acqua & Job, 1998). Contrary to the previous findings, the present study shows that, although silhouettes were identified with surprising good accuracy in the reality-decision task, effects of object identity on the elongation-decision task were negligible.