Structural patterns existing in language can be exploited for implicit prediction of sequences in speech and visual input via a process termed statistical learning (SL). Despite extensive examination of SL in dyslexia, whether SL problems arise from modality-constrained learning processes or from global learning processes is still unknown, nor is it clear how SL can be supported. Purpose: The present study used the triplet paradigm to explore SL among young adults with dyslexia and among typical readers across auditory and visual modalities and tested whether information from one sensory modality can assist SL in a different sensory modality. Method: Participants performed auditory and visual SL tasks under conditions in which a consistent visual/auditory cue respectively accompanied the auditory/visual triplets or under conditions in which no cross-modal information was presented. Results: SL performance was poorer in the dyslexia group than among typical readers across visual and auditory modalities. Furthermore, both groups improved their SL abilities under conditions in which cues were consistent with triplet boundaries compared to under conditions lacking cross-modal information Conclusions: These findings suggest that SL impairments observed in dyslexia stem from a domain-general deficiency and that cross-modal information can be recruited to support SL in dyslexia.