This quasi-experimental study investigated how explicit instruction about multiple modes of representation (MMR) impacted grades 7 (n = 61) and 8 (n = 141) students' learning and multimodal use on end-of-unit assessments. Half of each teacher's (n = 3) students received an intervention consisting of explicit instruction on MMR in science discourse, in addition to regular science instruction enhanced by a focus on MMR; comparison groups of students received regular science instruction. Three ordinary least squares regression models used student demographic variables and whether or not students received the intervention to predict students' (a) gain scores on end-of-unit tests, (b) voluntary use of embedded MMR on unit tests, and (c) retention of science knowledge as measured by a state end-of-level criterion-referenced assessment. Analyses showed that explicit instruction on MMR did not make a significant impact on student gain scores, the amount of embeddedness on unit tests, or end-of-level scores. However, Models 2 and 3 showed Hispanics and females used MMR more on end-of-unit tests than Whites or males, respectively, whether or not they received the intervention. Hispanics and females scored lower than Whites or males, respectively, on end-of-level, multiple-choice assessments. Implications for classroom teachers and educational researchers in relation to these underserved populations are discussed.