Procedural learning is thought to play a crucial role in language and literacy acquisition through the extraction of linguistic regularities from the sensory input. Aligning with this, a procedural deficit is hypothesized to be causally implicated in neurodevelopmental disorders such as dyslexia and developmental language disorder. In this thesis, we comprehensively examined the role of procedural memory in the serial reaction time task (SRTT) in language/literacy in typical and atypical populations. We began by establishing, and potentially improving, the reliability of the SRTT in typically developing adults by examining the impact of the similarity between sequences at test and retest, the addition of an extra session, and the inclusion of an interstimulus interval. Despite numerical improvements, no experimental manipulation resulted in adequate test-retest reliability. This was further confirmed in a meta-analysis that assessed the reliability of the SRTT. Alongside the careful examination of the psychometric properties of the SRTT, the relationship between procedural learning and language/literacy was reported, with no support for associations between these measures. This was also replicated in a second meta-analysis in typical and atypical populations. Additionally, the performance of adults with and without dyslexia on the SRTT was contrasted across three sessions. This study represented the first attempt to assess the reliability of the SRTT in a disordered population. Again, the test-retest reliability of the SRTT was well-below adequate psychometric standards. There was no evidence for a procedural learning impairment in the dyslexic participants in any session, and there was only a moderate relationship between nonword repetition and procedural learning in this group. Nonetheless, procedural learning and attention were consistently correlated across experiments. To conclude, these findings do not rule out the possibility that procedural learning is involved in language/literacy development and disorders; instead, they highlight the need for more reliable measures and more testable hypotheses.