Four experiments explored the composition and stability of internal orthogr
aphic representations of printed words. In three experiments, subjects were
presented on successive occasions with words that were consistently spelle
d correctly or were consistently misspelled. On the second presentation, su
bjects were more likely to judge both kinds of words as correctly spelled t
han on the first presentation, suggesting that their preexperimental orthog
raphic representations had been altered to match what they had seen on the
first presentation. However, only misspellings that were consistent with th
e correct phonology were accepted; spellings that altered the phonology wer
e rarely accepted, suggesting that some parts of the orthographic represent
ation are less stable than others. Also, subjects' reliance on orthographic
vs. phonological memory when judging a word's spelling was affected by the
kinds of other misspellings in the list. Lists that contained some phonolo
gically implausible spellings for real words (e.g., *assostance) induced su
bjects to rely more on phonological plausibility when judging the correctne
ss of other words in the list and less on orthographic memory. An individua
l grapheme in an internal orthographic representation was unstable when the
re were many phonologically acceptable alternatives for it. The results are
contrary to the view that the strength of an internal representation is un
iform across all its graphemes and is a function only of visual experience
with the printed form. Results were interpreted in the context of a theory
that considers spelling knowledge to be a by-product of the reading process
, a process that involves phonological analysis.