The processing required to categorize faces and animals is not only rapid but also remarkably resistant to inversion. It has been suggested that this sort of categorization performance could be achieved using the global distribution of orientations within the image, which interestingly is unchanged by inversion. Here, we presented subjects with two natural scenes at 16 different orientations that were simultaneously flashed in the left and right hemifield and we asked them to make a saccade to the side containing an animal. We report that human performance is surprisingly rotation invariant as reaction times were similar and accuracy remarkably stable across orientations. The results imply that this form of rapid object detection could not depend on the global distribution of orientations within the image. One alternative is that subjects are instead using local combinations of features that are diagnostic for the presence of an animal.