Miettinen and Cook (Am J Epidemiol 1981;114:593-603) postulate that confounding is a result of exposure-strata dependence in the joint source population, irrespective of the effective parameter used. Boivin and Wacholder (Am J Epidemiol 1985;121:152-8) postulate that confounding is dependent on the additive structure of the crude effect parameter. Both of these conflicting principles are at variance with the familiar definition in terms of parameter bias. It is argued that the change-in-parameter definition is the only fundamental definition.