A quantitative synthesis of multivariate studies in science education during the past ten years yields evidence to support a model of achievement in science very much like that currently proposed in the literature of expert problem-solving. Despite previous claims, spatial visualization and verbal ability predict additional variance in achievement beyond that which they share with general intelligence. Both are also components of most theories of expertise. Cognitive ability, scientific reasoning, and prior knowledge all add further variance to the regression equation, and are demonstrably separate in their contributions to scientific achievement. Quantitative ability does not appear to be uniquely separate from cognitive ability. These results suggest an instructional program that focuses more on domain-specific skills and knowledge than on generalized and problem-solving. (Author)