Visual scanning and exploration of natural scenes not only depends on specific objects, but also on local features. Models of spatial attention propose that features such as orientation or colour are processed pre-attentively and in parallel. According to these models attention interferes at a later stage, where features are combined into a representation of visual saliency. Saliency is a good predictor of ocular fixations during scanning of static pictures. Here, we tested whether fixations of patients with left spatial neglect, hemianopia, or both can be predicted based on local image content. Participants were asked to freely scan natural images while saccades and ocular fixations were registered. Hemianopic patients produced a similar distribution of fixations and relied similarly on picture saliency as healthy controls. In contrast, neglect patients looked to image regions with increased saliency and higher local orientation and intensity thresholds on the neglected side of space. The reliance on increased saliency during visual exploration was predicted by damage to subcortical regions interconnecting the inferior parietal and lateral premotor cortex. These findings suggest that spatial neglect leads to a combined attentive and pre-attentive deficit in the processing of saliency and feature information.