Error management theory is a theory of considerable scope and emerging influence. The theory claims that cognitive biases do not necessarily reflect flaws in evolutionary design, but that they may be best conceived as design features. Unfortunately, existing accounts are vague with respect to the key concept of bias. The result is that it is unclear that the cognitive biases that the theory seeks to defend are not simply a form of behavioral bias, in which case the theory reduces to a version of expected utility theory. We propose some clarifications and refinements of error management theory by emphasizing important distinctions between different forms of behavioral and cognitive bias. We also highlight a key assumption, that the capacity for Bayesian beliefs is subject to constraints. This assumption is necessary for what we see as error management theory's genuinely novel claim: that behavioral tendencies to avoid costly errors can rest on systematic departures from Bayesian beliefs and that the latter can be adaptive insofar as they generate the former.