This study of conversation was undertaken to test experimentally conclusions reached in earlier work with deaf primary school children and pre-school hearing children. Both of these studies revealed significant negative correlations between a measure of teacher control of the conversations and measures of children's initiative and loquacity. This study was designed to investigate the direction of causality in these correlations. Teachers were asked to change their conversational styles in specific ways with the same pairs of children. On each of five occasions they were to bias their conversations towards one of five "levels of control"--enforced repetitions, two-choice questions, wh-type questions, personal contributions and phatics. The results show that as teachers change style, their children follow them with the predicted changes in initiative and mean length of turn.