Listeners implicitly use statistical regularities to segment sound input into meaningful units, e.g., a continuous speech stream into separate phonemes. This innate statistical learning ability likely plays a crucial role for language development. In particular, the efficient formation of representations for distinct phonemes from the speech input is an important foundation for reading and spelling development. Thus, it was suggested that literacy deficits in dyslexia are grounded in poor statistical learning abilities. In my talk, I will present two parts of an ongoing project that combines EEG and computational modelling to study the neural and computational mechanisms behind the link between statistical learning and literacy skills, which so far remains poorly understood. First, I will present the results of an EEG study that investigated neural tracking of single elements, i.e., individual tones, and statistical chunks, i.e., fixed triplets of tones that always occurred together, within a continuous tone sequence in adults with and without dyslexia. Individuals with dyslexia showed an altered pattern of neural tracking, and overall stronger tracking of statistical chunks (relative to single elements) was associated with better spelling skills. Second, I will introduce a modelling study that aims to model individual learning trajectories while participants learn the statistical triplet structure of the sequence to unravel systematic differences between individuals with and without dyslexia. Together, these data help to understand the link between statistical learning and literacy skills, which provides an important basis for the development of novel intervention approaches to ameliorate phonological deficits in dyslexia.