Background/Context: How do we account for the persistence of below-average math test score performance among California Hispanics who are fluent in English, as well as Spanish-dominant English learners? Recent studies have attributed the problem to an overly rigid focus on "what works" in curriculum and fluency in English to the veritable neglect of the social components of teaching and learning--particularly caring. Purpose/Objective/Research Question/Focus of Study: We investigated Hispanic elementary student perceptions of teacher caring in relation to their math self-efficacy and math test performance, and we specify the sequence of the relationship: Caring teachers bolster student self-efficacy in math, which in turn bolsters math test scores. Moreover, we sought to examine whether the meditational relationships among the variables were moderated by English language proficiency. Research Design: Our correlational/comparative analyses were based on longitudinal data for 1,456 Hispanic students nested in 84 fifth- or sixth-grade classrooms in the spring of 2007. Students were either fluent English speakers (EFs, n = 799) or English learners (ELs, n = 657). We secured student self-report measures of teacher caring and math self-efficacy using the Student Motivation Questionnaire, and scores from the California Standards Test for Mathematics served as the primary dependent variable. While controlling for background variables, prior math achievement, and prior math self-efficacy where appropriate, we employed a well-known framework and a series of multilevel regression models to examine our hypothesis of moderated mediation. Conclusions/Recommendations: For all study participants, caring teachers bolstered can-do attitudes in math, which in turn positively impacted math test scores. We identified two principal differences, however, in support of our hypothesis of moderated mediation that indicate that the total effect of teacher caring is larger among ELs. First, the magnitude of the direct link between teacher caring and math self-efficacy was more pronounced among ELs. Second, teacher caring was only partially mediated by math self-efficacy for ELs, whereas for EFs, the positive influence of teacher caring on math scores was completely mediated by math self-efficacy. Several issues come to light when the literature on how communication across cultural and language barriers impacts perceptions of caring is examined concurrently with our findings. Among them is the deemphasis of bilingual ability in California's recent mandate for more authorizations to teach ELs, which may create a barrier to fostering caring teacher-student and teacher-parent relations for Hispanic EFs and especially Hispanic ELs, whose math achievement would otherwise stand to gain.