Children coming from homes where English is not their first language constitute a significant and increasing proportion of classrooms worldwide. Providing these English-language learners (ELLs) with equitable assessment opportunities is a challenge. We analyze the performance of 485 students, both English-native-speakers (ENSs) and ELLs, across five schools within the United Kingdom in the 7-11-year age group on standardized summative Science assessment tasks. Logistic regression with random effects assesses the impact of English-language proficiency, and its interactions with question traits, on performance. Traits investigated were: question focus; need for active language production; presence/absence of visuals; and question difficulty. Results demonstrated that, while ELLs persistently performed more poorly, the gap to their ENS peers depended significantly upon assessment traits. ELLs were particularly disadvantaged when responses required active language production and/or when assessed on specific scientific vocabulary. Presence of visual prompts did not help ELL performance. There was no evidence of an interaction between topic difficulty and language ability suggesting lower ELL performance is not related to capacity to understand advanced topics. We propose assessment should permit flexibility in language choice and production type for ELLs with low English language proficiency; while simultaneously recommend subject-specific teaching of scientific language begins at lower stages of schooling.