There is growing interest in children’s sensitivity to orthographic probabilistic constraints governing well-formed letter sequences in print. A few studies have demonstrated that young children’s spellings conform to some untaught orthographic restrictions and have postulated that distributional statistical learning processes may underlie this ability. However, there are no studies investigating whether similar patterns can be learnt under experimental conditions whereby participants are not instructed to learn. Therefore, an incidental learning task was used in this thesis to investigate whether novel constraints on letter positions and letter contexts can be exploited and used in a subsequent legality discrimination task by 7-year-old children and adults. Results indicated that (a) novel letter positions and contexts were reliably learnt by children and adults. (b) Adults were, by and large, superior learners. (c) Children’s and adults’ ability to learn was similarly affected by pattern complexity. These findings confirm the statistical nature of children’s sensitivity to general properties of their orthography. The next question addressed was whether implicit sensitivity to frequency-based information that is widely embedded in written language (allowable letter positions; probable/improbable letter pairs/triplets and larger units) is impaired in dyslexia. To this aim, variants of an implicit artificial grammar learning task were used among groups of dyslexic adults and skilled adult readers. It was demonstrated that (a) in letter versions of the task, most aspects of dyslexics’ performance were spared relative to skilled readers’ performance. (b) Importantly, in a nonlinguistic version, dyslexics’ performance matched that of their skilled reader counterparts in every aspect. These results contrast with some previous studies of implicit learning in dyslexia and challenge the claim that a general learning deficit contributes to reading and spelling disability.