Conjectural variation models have been criticized because firms are ‘right for the wrong reasons’, and because possible equilibria are abundant Lately several papers have attempted to meet these difficulties by some condition of consistency or rationality on the conjectural variation This paper shows that stronger versions of such conditions lead into paradoxes, while the weaker forms are of little use in solving the conceived problems The original Cournot assumption actually satisfies consistency enterions The problems with consistent conjectures illuminates fundmental problems with too demanding information assumptions like perfect foresight Such assumptions may create non-trivial structures of self-reference.