Most research on linguistic features of test items negatively impacting English language learners' (ELLs') performance has focused on lexical and syntactic features, rather than discourse features that operate at the level of the whole item. This mixed-methods study identified two discourse features in 162 multiple-choice items on a standardized science assessment. After analyzing the frequency of features, we correlated the presence of these features with values of Differential Item Functioning (DIF) favoring non-ELLs over ELLs. Next, we analyzed 52 interviews with ELLs to examine the interactions between students and items with and without these features. Our results indicate that these two discourse features negatively impacted ELLs' item performance and item comprehension, indicating they should be avoided when developing test items.