Visual processing is a language-agnostic window into heterogeneity in early reading development.

Reading disabilities impose profound impacts on educational and socioeconomic outcomes worldwide, yet identifying reliable early predictors remains a significant challenge. For five decades, the role of visual processing in reading development has been contested-largely due to small, homogeneous samples and a lack of research designed to capture heterogeneity in reading development. In this study, we administered theory-driven, carefully validated measures of visual processing-including multi-element processing of letters and pseudo-letters and global motion coherence-to a large, socioeconomically and linguistically diverse cohort of kindergarten and first-grade children in California public schools. A subset of children received language-based assessments and reading outcomes along with the visual measures (n ∼ 400). Using latent profile analysis, we identified subgroups invisible to conventional screening: children with strong language skills but poor visual-processing abilities who exhibited persistent reading difficulties a year later and another subgroup of children who leveraged strong visual-processing abilities to achieve reading success despite weaker language abilities. Visual-processing measures proved equitable across children's socioeconomic status and home language, independently accounting for 12%-16% of variance in reading outcomes and emerging as consistently important predictors of reading risk in machine learning models. Integrating measures of rapid visual processing, especially multi-element processing, into early identification offers a path toward mechanistically informed personalized interventions that leverage children's specific strengths.