A recent critique of using teachers' test score value-added (TVA) is that teacher quality is multifaceted; some teachers are effective in raising test scores, others are effective in improving long-term outcomes This paper exploits an institutional setting where high school teachers are randomly assigned to classes to compute multiple long-run TVA measures based on university schooling outcomes and high school behavior. We find substantial correlations between test scores and long-run TVA but zero correlations between these two TVA measures and behavior TVA. We find that short-term test-score TVA and long-run TVA are highly correlated and equally good predictors of long-term outcomes.