This study examined the hypothesis by Chomsky and Halle (1968) and C. Chomsky (1970) that English spelling corresponds to an underlying lexical level of representation rather than to a surface phonetic level, i.e., pronunciation. And that learning to read involves learning the spelling-lexical level correspondences. This hypothesis was tested by comparing the performance of 87 second and third grade Ss on a paired-associate and reading task composed of a morphologically related list of word pairs (REL), e.g., hide—hid and an unrelated (UNR) list of word pairs, e.g., pine—pin. Both sets of pairs exhibit vowel alternations but are stored differently in the internal lexicon. It was predicted that REL would be superior on the tasks to UNR for all Ss and that the difference would be greater for the better readers over poorer readers and the grade 3 Ss over the grade 2 Ss. A multivariate and univariate analysis of the difference scores between the REL and UNR lists on three criterion measures failed with one exception to confirm these predictions. This study did not offer much support for the predictions of transformational theory about reading acquisition.