This paper analyzes the effect of limited information and estimation risk on optimal portfolio choice when the joint probability distribution of security returns is multivariate normal and the underlying parameters (means and variance-covariance matrix) are unknown. We first consider the case of limited, but sufficient information (the number of observations per security exceeds the number of securities or the prior distribution of the underlying parameters is “sufficiently” informative). We show that for a general family of conjugate priors, the admissible set of portfolios, taking estimation risk into account, may be obtained by the traditional mean-variance analysis. As a result of estimation risk the optimal portfolio choice differs from that obtained by traditional analysis.