Children with a diagnosis of autism and typically developing children were given two variations of the Navon task (Navon, 1977), which required responding to a target that could appear at the global level, the local level, or both levels. In one variation, the divided attention task, no information was given to children regarding the level at which a target would appear on any one trial. In the other, the selective attention task, children were instructed to attend to either the local or the global level. Typically developing children made most errors when the target appeared at the local level whereas children with autism made more errors when the target appeared at the global level in the divided attention task. Both groups of children were quicker to respond to the global target than the local target in the selective attention task. The presence of normal global processing in the children with autism in one task but not in the other is discussed in terms of a deficit in mechanisms that inhibit local information in the absence of overt priming or voluntary selective attention to local information.