In many empirical investigations, particularly in fields such as economic policy, personalized medicine, and direct marketing, estimating a treatment effect on a response is a primary goal. Typically, we would like to select a statistical model based on sample data to estimate conditional treatment effects. However, the average from the candidate models often provides a more accurate estimate than the selection of a single candidate model. This paper proposes a new weight choice for model average estimate of the treatment effect conditional on covariates. We prove that our new model average estimator is asymptotically optimal in the sense of achieving the lowest possible squared error. In a simulation experiment we show that the proposed estimator compares favorably with those based on AIC and BIC weights. We apply the averaging method to evaluate the effect of the labor market program.