Following Kuhn's main thesis according to which theory revision and acceptance is always paradigm relative, I propose to outline some possible consequences of such a view. First, asking the question in what sense Bayesian decision theory could serve as the appropriate (normative) theory of rationality examined from the point of view of the epistemology of theory acceptance, I argue that Bayesianism leads to a narrow conception of theory acceptance. Second, regarding the different types of theory revision, i. e. expansion, contraction, replacement and residuals shifts, I extract from Kuhn's view a series of indications showing that theory replacement cannot be rationalized within the framework of Bayesian deciston theory, not even within a more sophisticated version of that model. Third, and finally, I will point to the need for a more comprehensive model of rationality than the Bayesian expected utility maximization model, the need for a model which could better deal with the different aspects of theory replacement. I will show that Kuhn's distinction between normal and revolutionary science gives us several hints for a more adequate theory of rationality in science. I will also show that Kuhn is not in a position to fully articulate his main ideas and that he well be confronted with a serious problem concerning collective choice of a paradigm.