We argue, from an extensive literature review, that in the vast majority of research settings, biases in alternative expected-return proxies (ERPs) are irrelevant. Therefore, in most settings, the choice between alternative ERPs should be based on an evaluation of their relative measurement-error variances. We develop a parsimonious evaluation framework that empirically estimates a given ERP’s cross-sectional and time-series measurement-error variances. We then apply this framework to five classes of firm-level ERPs nominated by recent studies, including factor-based ERPs from finance and implied costs of capital (ICC) estimates from accounting. Our analyses show ICCs are particularly useful in tracking time-series variations in expected returns. We also find broad support for a “fitted” or “characteristic-based” approach to ERP estimation.