The regularity of spelling-to-sound correspondences has played a major role in reading research (Coltheart, 1978; Glushko, 1979; Seidenberg & McClelland, 1989; Coltheart & Rastle, 1994). However, despite its critical importance, this property of words has never received a standard workable definition. Regularity is usually defined as conformity to the spelling-to-sound rules as given by Venezky (1970), but important variations exist (see Henderson, 1985b). The purpose of this work was to investigate regularity in more details. In a first study, I developed a graded meaure of Grapheme-Phoneme Association Strength (GPAS) based on computational analyses of spelling-sound correspondences in English disyllabic words. The psychological validity of the GPAS measure was verified in an empirical study. Both the GPAS variable and a subjective estimate of the letter-sound regularity were significantly related to immediate (but not delayed) naming times. In a regression study, I confirmed the finding of a regularity effect at the level of grapheme-phoneme correspondences (GPCs) on two sets of monosyllabic and disyllabic words and observed that the effect was better explained by the degree of regularity than by a regular-exception dichotomy. The degree of regularity was attributed to two highly correlated factors, the GPAS measure of graded GPC-regularity or the number of irregular GPCs in the word. The results are discussed in the context of current models of word reading.