Background: Remarkably few studies have investigated the nature and origin of learning difficulties in children with cerebral palsy (CP). Aims: To investigate math achievement in terms of word-problem solving ability in children with CP and controls. Because of the potential importance of reading for word-problem solving, we investigated reading as well. Sample: Children with CP attending either special (n = 41) or mainstream schools (n = 16) and a control group of typically developing children in mainstream schools (n = 16). Method: Group differences in third grade math and reading, controlled for IQ, were tested with analyses of co-variance (ANCOVAs). Hierarchical regression was used to investigate cognitive correlates of third grade math and reading. Predictors included verbal and non-verbal IQ measured in first grade, components of working memory (WM) and executive function (EF) measured in second grade, and arithmetic fact fluency and reading measured in third grade. Results: Children with CP in special schools performed significantly worse than their peers on word-problem solving and reading. There was a trend towards worse performance in children with CP in mainstream schools compared to typically developing children. Conclusions: Impairments of non-verbal IQ and WM updating predicted future difficulties in both word-problem solving and reading. Impairments of visuospatial sketchpad and inhibition predicted future word-problem, but not reading difficulty. Conversely, deficits of phonological loop predicted reading but not word-problem difficulty. Concurrent arithmetic fact fluency and reading ability were both important for word-problem solving ability. These results could potentially help to predict which children are likely to develop specific learning difficulties, facilitating early intervention. (Contains 4 tables.)