The present study used social cognitive career theory (SCCT; Lent, Brown, & Hackett, 1994) to predict the math/science goal intentions of a sample of low-income prospective first-generation college students (N = 305). Structural equation modeling was used to test a model depicting relationships between contextual (i.e., social class, learning experiences, proximal supports and barriers) and person-cognitive (i.e., self-efficacy, outcome expectations, interests, goals) variables as hypothesized in SCCT and based on previous literature on low-income first-generation college students. Results indicated that the hypothesized model provided the best representation of the data. All paths in the model were statistically significant, with the exceptions of paths from self-efficacy to goals, outcome expectations to interests, and perceived barriers to self-efficacy. Bootstrapping procedures revealed that the relationships between social class, self-efficacy, and outcome expectations were mediated through learning experiences. Furthermore, the relationship between social supports and goals was mediated by self-efficacy and interests and the relationships between self-efficacy, outcome expectations, and goals were mediated by interests. Contrary to hypotheses, the relationship between barriers and goals was not mediated by self-efficacy and interests. The hypothesis that proximal contextual supports and barriers would moderate the relationship between interests and goals was not supported. The final model explained 66% and 55% of the variance in math/science interests and goals, respectively. Implications for future research and practice are discussed. (Contains 4 tables and 1 figure.)