This study examines how linguistic features in the Swedish Grade 8 Science items from TIMSS 2011 correlate with results from different groups of students. The language use in different science subjects is analysed using four characteristic meaning dimensions of science: Packing, Precision, Presentation of information and Personification within the text, as well as using conventional measures of readability and information load. For each subject, one or more of the meaning dimensions show statistically significant correlations with students’ performances. Many meaning dimensions shown to influence high performers’ results are not influencing low performers, and vice versa. The use of meaning dimensions is shown to be an enriching complementary method for analysing language use in science, as it links language use in items with student performances where conventional measures do not. Unexpected findings are that placing items in everyday contexts lowers the likelihood of them being answered correctly, and that skilled readers are aided by high precision in items whereas less skilled readers seems unaffected.