This paper presents new evidence on the returns to schooling based on an interactive fixed effects framework that allows for multiple unobserved skills with associated prices that are potentially time-varying. Skills and prices are both allowed to be correlated with schooling. The modeling approach can also accommodate individual-level heterogeneity in the returns to schooling. The framework thus constitutes a substantive generalization of most existing approaches that assume ability is unidimensional and/or returns are homogeneous. Our empirical analysis employs a unique panel dataset on earnings and education over the period 1978-2011 based on respondents from the Survey of Income and Program Participation (SIPP) linked with tax and benefit data from the Internal Revenue Service (IRS) and Social Security Administration (SSA). Our preferred specification yields a point estimate of the average marginal returns to schooling of about 2.8-4.4 percent relative to ordinary least squares and two stage least squares estimates which lie in the range 7.7-12.7 percent. A decomposition of the aggregate least squares bias shows that the omitted ability component is responsible for a larger fraction of the bias relative to the heterogeneity component. Finally, our heterogeneity analysis suggests larger returns for individuals born in more recent years, the presence of sheepskin effects, and considerable within-group heterogeneity.