Correlational studies have demonstrated detrimental effects of exposure to a mismatch between a non-standard dialect at home and a mainstream variety at school on children’s literacy skills. However, dialect exposure often is confounded with reduced home literacy, negative teacher expectation and more limited educational opportunities. To provide proof of concept for a possible causal relationship between variety mismatch and literacy skills, we taught adult learners to read and spell an artificial language with or without dialect variants using an artificial orthography. In three experiments, we confirmed earlier findings that reading is more error-prone for contrastive words, i.e. words for which different variants exist in the input, especially when learners also acquire the joint meanings of these competing variants. Despite this contrastive deficit, no detriment from variety mismatch emerged for reading and spelling of untrained words, a task equivalent to non-word reading tests routinely administered to young school children. With longer training, we even found a benefit from variety mismatch on reading and spelling of untrained words. We suggest that such a dialect benefit in literacy learning can arise when competition between different variants leads learners to favour phonologically mediated decoding. Our findings should help to assuage educators’ concerns about detrimental effects of linguistic diversity.