Following a stroke, patient TL presented with a pure alexia that also displayed the features of phonological alexia (poor pseudoword reading, part-of-speech effect). This pattern was absent, however, when words were spelled aloud to TL. We propose that, with visual input, orthographic word forms are only weakly activated in TL. This weak activation succeeds in activating strongly represented semantic code—such as concrete nouns—but not weakly represented semantic cod—such as functor words—resulting in a part-of-speech effect. When words are spelled aloud, the input is sufficient to activate the orthographic word form strongly, and all words are identified. Most pure alexia patients can spell written words aloud to themselves (hence they are often called letter-by-letter readers), and consequently do not show symptoms of semantic reading. TL, whose letter-naming abilities are poor, cannot resort to this alternative strategy, and so the consequences of a weakly activated orthographic word form are revealed.