Johnson-Laird and Byrne (1991) present a theory of conditional inference based upon the manipulation of mental models. In the present paper, the theory is critically examined with regard to its ability to account for psychological data, principally with respect to the rate at which people draw the four basic inferences of modus ponens (MP), denial of the antecedent (DA), affirmation of the consequent (AC) and modus tollens (MT). It is argued first that the theory is unclear in its definition and in particular with regard to predictions of problem difficulty. Clarification and specification of principles are consequently provided here. Next, it is argued that there are a number of phenomena in the conditional reasoning literature for which the theory cannot account in its present form. Specifically, (a) the relatively frequency of DA and AC inferences on affirmative conditional is not as predicted by the theory, (b) differences occur between inferences on if then and only if rules beyond the capacity of the theory to explain and (c) there is no account of the "negative conclusion bias" observed when negated components are introduced into the rules. A number of revisions to the mental model theory of conditional reasoning are proposed in order to account for these findings.