Research documents an income-based achievement gap in mathematics, yet children from lower-income backgrounds do not lag behind their more advantaged peers in high-level social reasoning tasks. The purpose here was to investigate whether modifying mathematics word problems to make them more socially based would impact the mathematics performance and/or mathematics self-efficacy of lower- versus higher-income children. Research questions regarding (1) the relative difficulty of symbolic equations versus word problems, (2) the impact of socially modifying word problems on children’s accuracy and self-efficacy, and (3) the relation between children’s mathematics performance and mathematics self-efficacy were explored. Participants were 164 5th graders. Children completed a mathematics problem-solving test comprised of multiplication problems representing four different problem formats (two social, two abstract). Three types were word problems, and one was a symbolic (abstract) presentation. The three word problem types were everyday activity (social), social-cognitive (social), and traditional textbook (abstract). Participants also completed a mathematics self-efficacy measure. Children performed better on symbolic problems than on any of three word problem types. The lower-income group performed better on innovative social-cognitive word problems than on decontextualized word problems. Word problem variations did not have an effect for the higher-income group. Overall, mathematics self-efficacy was shown to predict mathematics performance. While problem format is only one aspect of a highly complex instructional system, findings suggest that capitalizing on social-cognitive strengths in mathematics may be valuable for improving the academic achievement of lower-income children.