A patient with bilateral parietal damage, and Balint's syndrome, named visual letters. These were presented individually or within four-letter strings. Solitary letters were identified very accurately. In the case of strings, more letters were correctly reported for words than for nonwords, and more for pronounceable than for unpronounceable nonwords. When required to read words as a whole, performance was better than predicted by letter-reports. These results extend the object-based limitation apparent in Balint's syndrome to the case of reading. The component letters of a string benefit when they form a familiar global object, rather than requiring representation as multiple separate objects. The patient occasionally made homophonic errors when listing the letters in a visual word. This suggests an attempt to bypass visual simultanagnosia by treating the string as a single object, deriving a holistic phonological code for it, and then decomposing this into component letters via spelling rules.