In this paper we propose an analysis of the semantic structure of modal sentences in English. Central to this analysis is the notion of ‘speech act’. The notion of speech act derives principally from the work of J. L. Austin (cf. especially Austin, 1962). Austin is particularly concerned with the use of sentences like I promise to go, I bet you sixpence and I name this ship ‘Queen Elizabeth’. One of his main purposes is to show that it is a mistake to class these sentences with sentences like John is running as statements. To utter the sentence I promise to go is not to make a statement about or to report some inner condition of mind or consciousness. It is to perform the act of promising, just as to say I bet you sixpence is to perform the act of betting. Indeed, in the case of these kinds of acts (promising, betting, naming, etc.) it is difficult to see how one could engage in them without using sentences of this kind. Austin calls verbs like promise, bet and name ‘performative’ verbs. It follows from this account that verbs like say, state and assert are also performative verbs. So that the mistake involved in classing I promise to go with John is running is not that in uttering the first sentence one is performing an act and in uttering the second one is not (ignoring the uninteresting fact that the act of vocalizing is common to both), but that the speaker is performing a different act in each case—in the one case promising and in the other stating. The fact that in the case of the utterance John is running the performative verb state does not occur, so that what Austin terms its ‘illocutionary force’ is not explicitly marked in the way in which the utterance I promise to go is explicitly marked as having the illocutionary force of a promise by the words I promise, while giving rise to certain difficult problems,2 does not affect Austin's main point, which is that a complete account of the meaning of a sentence cannot be restricted to semantic analyses as these are usually understood and that they must be extended to include information about the kind of speech act involved in uttering the sentence – that is, its illocutionary force.