Recently, Allan Gibbard has argued that there is a semantic difference between indicative and subjunctive conditional sentences. His arguments, which derive in part from the work of E.W. Adams (see [1] and [2] for example), are presented in detail in [3]. They have proved persuasive enough to dislodge Stalnaker (at least temporarily) from his previous thesis that the difference between indicative and subjunctive conditionals is pragmatic, rather than semantic. I want to show that Stalnaker's earlier position, as developed in [10], [11], and [12], can be defended against Gibbard's attack.' The philosophical point of this exercise is to show that 'if' need not be treated as semantically ambiguous. At the same time, we will find that the usual division of 'if-then' sentences into subjunctives and indicatives needs to be refined. Stalnaker's theory of conditionals is developed in the possible worlds framework: a proposition is thought of as a way of discriminating between possible worlds. Thus, a proposition is a subset of the set of all possible worlds. If A and B express propositions, the conditional 'If A then B1 (or 'A > B' ) also expresses a proposition.2 Which proposition a conditional expresses is determined by a selection function. This function takes as argument a pair (i,p), where i is a possible world, and p is a proposition. If p is a non-empty proposition, then f(i,p)Ep; if p is empty, then f(i,p) takes some default value which is not a possible world (say, the empty set). Intuitively, f(i,p) is the world in p which is most like i; the conditional FA >B' is true at a world i just in case f(i,p)Eq, where A and B express p and q respectively. If A expresses the impossible proposition (the empty set), then rA >BW is true, for any