There is substantial evidence that serial entrepreneurs outperform de novo entrepreneurs. But is this positive association between prior experience and performance the result of learning by doing or of selection on ability? This paper proposes a strategy that combines the fixed-effects model and IV estimations to distinguish empirically selection effects from learning. Using panel data from the NLSY79, I find that selection on ability is the more important determinant of serial business formation and the early performance of new businesses. In contrast, the effects of learning by doing are apparent only when the analysis focuses on founding new startups in sectors closely related to entrepreneurs' previous ventures.