Abstract
Two recent overviews of costly signalling theory—Maynard-Smith and Harper ( 2003 ) and Searcy and Nowicki ( 2005 )—both refuse to count signals kept honest by punishment of dishonesty, as costly signals, because (1) honest signals must be costly in cases of costly signalling, and (2) punishment of dishonesty itself requires explanation. I argue that both pairs of researchers are mistaken: (2) is not a reason to discount signals kept honest by punishment of dishonesty as cases of costly signalling, and (1) betrays too narrow a focus on certain versions of costly signalling theory. In the course of so arguing, I propose a new schema for classifying signal costs, which suggests productive research questions for future conceptual and empirical work on costly signalling