This paper reports on secondary analysis of TIMSS 2003 data, based on a sound conceptual model, aiming to explain differential science achievement in South Africa from the perspective of educational effectiveness research. The conceptual framework was developed by refining existing school effectiveness models and including factors related to science achievement. The refined model integrated psychological and sociological aspects and reflected the multilevel-structure of schools. The model added resources and climate to the quality factors at class/school level. It was applied to the South African results of TIMSS 2003. Data from the student (n=8,952), teacher (n=255) and school questionnaires (n=255) were analysed in conjunction with achievement data by means of factor, reliability, correlation and multilevel analysis. The multilevel analysis revealed that at student level the strongest predictor of science achievement is attitude towards science. At classroom/school level, the strongest predictors are resource- and climate-related factors such as the safety in school, physical resources and class size. Factors at class/school level influenced performance more than student level factors with 59% of the total variance in science achievement occurring at class/school level. Such results indicate that the model developed is well suited to science education in developing countries.