This paper uses a semiparametric latent variable transformation model for multiple outcomes to examine the effect of education and maternal education on female multidimensional well-being and proposes a procedure to build a well-being index that is less susceptible to functional form misspecification. We model multidimensional well-being as an unobserved common factor underlying the observed well-being outcomes. The semiparametric methodology allows us to alleviate misspecification bias by combining multiple indicators into a latent construct in an unspecified, data-driven way. Using data from female participants of the 1974–2010 waves of the US General Social Survey, we find that education, intelligence, and maternal education contribute positively to multidimensional well-being. However, the effects of education and maternal education on female multidimensional well-being declined steadily between the mid-1970s and the 1990s, and have not rebounded since.