Visual search involves the matching of visual input to a "search template," an internal representation of task-relevant information. The present study investigated the contents of the search template during visual search for object categories in natural scenes, for which low-level features do not reliably distinguish targets from nontargets. Subjects were cued to detect people or cars in diverse photographs of real-world scenes. On a subset of trials, the cue was followed by task-irrelevant stimuli instead of scenes, directly followed by a dot that subjects were instructed to detect. We hypothesized that stimuli that matched the active search template would capture attention, resulting in faster detection of the dot when presented at the location of a template-matching stimulus. Results revealed that silhouettes of cars and people captured attention irrespective of their orientation (0°, 90°, or 180°). Interestingly, strong capture was observed for silhouettes of category-diagnostic object parts, such as the wheel of a car. Finally, attentional capture was also observed for silhouettes presented at locations that were irrelevant to the search task. Together, these results indicate that search for familiar object categories in real-world scenes is mediated by spatially global search templates that consist of view-invariant shape representations of category-diagnostic object parts.