The left mid-fusiform gyrus (lmFG) is thought to play a key role in reading. Previous studies have shown that the representation of real words in the lmFG is initially coarse but evolves through time, eventually allowing for individuation of even orthographically similar words. Whether this individuation of orthographic representations is unique to real words, which have learned phonological and semantic associations in addition to their visual form, and the degree to which it is supported through interactions with other nodes of the language network, is unknown. The current study probed these questions by showing participants orthographically similar real words, pseudowords, consonant strings, and false fonts while monitoring lmFG activity with either intracranial electroencephalography or magnetoencephalography (MEG). Neural decoding analyses in each modality replicated the previous findings regarding the dynamics of dissimilar and similar real word decoding. In contrast, while dissimilar pseudowords, consonant strings, and false-fonts could be decoded, no individuation of these stimuli was seen in lmFG activity; that is, similar pseudowords, consonant strings, and false-fonts could not be decoded either earlier or later in time. MEG functional connectivity analysis demonstrated significantly higher connectivity between the lmFG and both early visual cortex and the left anterior temporal lobe during the individuation of real word representations in the lmFG. These results suggest that the individuation of word representations in the lmFG is catalyzed by stored knowledge about word forms that emerges from network-level interactions with anterior regions of the temporal lobe.