Poor verbal skills in poor readers have long been reported in the literature. There have been many attempts to understand the interaction between poor verbal ability and poor verbal achievement. The methodological problems are considerable, including the measurement of verbal ability, which has been confounded by previous learning. A new reasoning test, the VESPAR, has been designed to measure novel problem solving and thus to be less reliant on acquired verbal skills. One hundred and seventy 14-year-olds completed the VESPAR, the Cognitive Abilities Test (CAT) and a single-word reading test. Overall, verbal scores were weaker than spatial scores. A subgroup of 38 pupils with particularly marked discrepancies between verbal and non-verbal CAT was identified. The especially discrepant pupils were matched with other non-discrepant pupils from the year group for either verbal or non-verbal CAT. The discrepant group's reading was at the same level as the matched verbal CAT group. However, the primary verbal ability of the discrepancy group, as measured on the VESPAR, was greater than the matched verbal CAT group. This raises the possibility that CAT- but not VESPAR-discrepant pupils may be at particular risk of under-achievement in the verbal domain.