In two experiments, subjects read passages of text and circled instances of a target letter under normal conditions or while engaged in articulatory suppression. In Experiment 1, subjects searching for the letter e made a disproportionately large number of errors on the word "the" and more errors when e occurred in unstressed than in stressed syllables of three-syllable words. In Experiment 2, subjects searching for the letter f made an exceedingly large number of errors on the word "of." Articulatory suppression significantly reduced the stress effect in the three-syllable words but did not reduce the tendency to make errors on "the" or "of," suggesting that phonological recoding is responsible for this effect of stress but does not influence the unitization processes of reading.In a study published in 1966, D. W. J. Corcoran noticed that subjects made an inordinate number of errors in detecting specific letters in the frequent and familiar word "the." Although Corcoran explained this finding by proposing that subjects skip over the word' 'the" because of its syntactic and semantic redundancy, Healy (1976) proposed that this finding was due to unitization. She hypothesized that readers tend to employ the largest reading units available to them. In familiar, highfrequency words, these units are presumably larger than single letters, and might, in fact, be the entire word. Therefore, subelements of these familiar word units, that is, letters, would be likely to be missed in a letter detection task. Healy's experiments supported this hypothesis; she found a bias toward errors in familiar versus unfamiliar words, even when syntactic and semantic redundancy was eliminated because only nouns, not function words, were used, and these words were scrambled in the text.Healy and Drewnowski (1983) extended this investigation and refined the unitization model in an experiment in which they showed that letter detection improved dramatically when test words were misspelled. It was clear that these words were being scanned, not skipped over, or else the misspellings would not have drawn the readers' attention to more of the target letters. Drawing on the interactive reading model proposed by McClelland and Rumelhart (1981) and the earlier work on automatic processing by LaBerge and Samuels (1974) and others, Healy and Drewnowski proposed that written text is processed in a parallel fashion at all levels that are available to the reader. Thus, processing may be occurring