Recent work by Richard Roll has challenged the worth of portfolio performance measures based on the capital asset pricing model. This paper demonstrates that Roll's conclusions are due to his focusing on a 'truly' ex-ante efficient index. Using a choice and information theoretic framework, we show that an appropriate index is efficient relative to the probabilities assessed by the 'market'. Residual analyses and portfolio performance tests, using such an index, yield meaningful results for a wide class of information structures. Roll's primary criticisms, however, relate to tests of the asset pricing model itself. We argue that these criticisms are vastly overstated. © 1979.