Background When we look at a visual scene our scanning eye movements are not random. We have previously compared the eye movements made by two patients with associative visual agnosia with those of healthy controls.¹ Despite being unable to recognise the images, the patients initially fixated the same locations as healthy individuals over the first four fixations, suggesting low-level *bottom-up* control of initial eye movements in patients and healthy controls.

Methods and Results Patient DF (female, age 55 years) exhibits a different form of agnosia—apperceptive agnosia.² Despite this she was able to recognise 13/15 scenes presented for 10 s each. By contrast to the patients with associative agnosia, DFs initial eye movements diverged significantly from those of healthy controls (p<0.05). We compared her fixation locations to those predicted by a bottom-up saliency model. There was a high similarity (p<0.01) between locations fixated by DF and the locations predicted by the saliency model on the basis of colour, but not orientation or edge density. Correlation of colour information with fixations was significantly higher for DF than for healthy controls, and—unlike control eye patterns—did not decrease over time.

Conclusions The two classic types of visual agnosia can be distinguished by examination of their initial scanning eye movements to real scenes.