Three monkeys performed a categorization task and a recognition task with briefly flashed natural images, using in alternation either a large variety of familiar target images (animal or food) or a single (totally predictable) target. The processing time was 20 ms shorter in the recognition task in which false alarms showed that monkeys relied on low-level cues (color, form, orientation, etc.). The 20-ms additional delay necessary in monkeys to perform the categorization task is compared with the 40-ms delay previously found for humans performing similar tasks. With such short additional processing time, it is argued that neither monkeys nor humans have time to develop a fully integrated object representation in the categorization task and must rely on coarse intermediate representations.