The Reciprocal Effects Model (REM) posits that academic self-concept and corresponding achievement measures are reciprocally related over time. Although there is considerable support for the REM based on short-term, narrowly focused educational accomplishments, little research evaluates the long-term implications of this reciprocal pattern of relations. Using large-scale longitudinal data from the U.S. Educational Longitudinal Survey (16,197 high school year 10 students from 751 schools, followed up through age 26), we found support for the REM (based on math self-concept [MS]) and achievement collected in years 10 and 12), with and without control for covariates (gender, socioeconomic status, school-average socioeconomic status, and reading achievement). We then extend this basic REM model, demonstrating long-term implications of the REM based on total, direct, and indirect (mediated) effects of Year 10 MSC and achievement on: end-of-high-school outcomes (units in mathematics, and highest level of mathematics based on actual school transcripts); postsecondary outcomes at age 26 (educational attainment, completion of the STEM credential, math courses, and advance math units); and long-term expectations (educational, occupational, and STEM job). The total (direct + indirect) effects of year 10 MSC and achievement were statistically significant for all subsequent outcomes, but their influence on postsecondary and long-term expectations were largely mediated by intervening variables. However, consistent with REM logic, MSC effects were mediated in part via math achievement, and math achievement effects were mediated in part via MSC. Hence, the extended REM design provides an ideal basis for evaluating the long-term effects of MSC.