Advanced and novice readers in 1st grade and older disabled readers were given nonword reading and spelling tasks. In addition, they practiced learning to read simplified phonetic spellings of 16 words for several trials (e.g., 'messenger' spelled MESNGR, 'stupid' spelled STUPD). Following Reitsma's (1983) procedure, three days later subjects read originally learned spellings mixed with altered spellings in which single letters were added or deleted or replaced by phonetically equivalent or non-equivalent letters (e.g., MESNGR altered to MESNJR; STUPD altered to STUP). Subjects' latencies to read original and altered spellings as the practiced words were measured. In ANCOVs equating the groups for reading level, disabled readers read as many nonwords and spelled as many words as beginning readers, indicating equivalent alphabetic knowledge. However, disabled readers took significantly more trials to learn to read the 16 target words, indicating deficient sight word learning processes. In the latency task, all three groups read original words faster than some types of modified spellings, indicating that subjects were reading the words by lexical access and that they were sensitive to some letter alterations. Beginning readers were affected by letter changes in medial as well as initial and final positions of words whereas disabled readers were affected only by initial and final letter alterations. From results we conclude that beginning readers acquire more complete grapho-phonic connections than disabled readers when they store sight words in lexical memory. Disabled readers' difficulties appear to arise from faulty word learning processes that impair the quality of lexical representations stored in memory.