The boundary paradigm was used to investigate individual differences in the extraction of lexical information from the parafovea in sentence reading. The preview of a target word was manipulated so that it was identical (e.g., "sped"), a higher frequency orthographic neighbor ("seed"), a nonword neighbor ("sted"), or an all-letter-different nonword ("glat"). Ninety-four skilled adult readers were assessed on measures of reading and spelling ability. The results showed that null effects of preview lexical status in the average data obscured systematic differences on the basis of proficiency and target neighborhood density. For targets from dense neighborhoods, inhibition from a higher frequency neighbor preview occurred among highly proficient readers, and particularly those with superior spelling ability, in early fixation measures. Poorer readers showed inhibition only in second-pass reading of the target. These data suggest that readers with precise lexical representations are more likely to extract lexical information from a word before it is fixated. The implications for computational models of eye movements in reading are discussed.