Maximizing expected utility is widely accepted in theoretical work, but extension to empirical models is more controversial. We state two theorems which outline conditions when mean-variance procedures such as quadratic programming can be used to maximize expected utility. These conditions are often assumed or satisfied in empirical studies.