We adapt the metric of Kandel and Stambaugh (1995) to evaluate linear asset pricing models. The “KS-ratio” criterion rates a model’s usefulness based on the mean portfolio return a mean-variance decision maker obtains for any variance choice by using the model for optimal portfolio decisions. It is equivalent to a cross-sectional GLS R-square criterion and to a measure of minimum distance between the asset and factor frontiers. We assess the KS-ratio compared to the HJ-distance and ad hoc goodness-of-fit evaluation criteria with simulated returns. We then apply the various criteria to evaluate nine prominent asset pricing models with actual data.