This paper argues for a new behavioural model of language facilitation. Current behavioural language programmes are critically analysed in terms of their poor generalizability to the natural language environment. More cognitively oriented developmental language research is cited as providing a relevant data base upon which to build behavioural language facilitation procedures. Such methods, which exploit the natural language environ ment and which emphasize child-initiated conversations, are more likely to facilitate appropriate language use which, in turn, is more likely to generalize. Practical examples of such methods are described. Finally, tentative implications of this for a behavioural theory of language teaching are proposed for consideration.