This paper presents three experiments designed to test Karl Popper's idea that scientists should try to falsify, or disconfirm, their hypotheses instead of verifying them. Previous studies indicate that subjects trained to falsify do no better on scientific problem-solving tasks than subjects trained to confirm. But these studies have all focused on individual subjects. Scientists frequently work in groups. To see how groups disconfirm, subjects in the first two studies reported here were run in groups of four. They were asked to solve a series of related problems based on ‘New Eleusis’, a task designed to model the ‘search for truth’.

In the first study, group members were either allowed to interact or told to work separately. This variable was crossed with a strategy (disconfirmatory vs. confirmatory) variable. There were no significant differences.

In the second study, only interacting groups were run, some methodological problems with the strategy variable were cleared up and a third strategy that combined elements of confirmation and disconfirmation was added. Disconfirmatory groups solved significantly more problems (72 per cent) than confirmatory (25 per cent) with combined (50 per cent) falling in between. Disconfirmatory groups also made significantly more attempts to falsify their hypotheses.

To see if the same results would hold for individuals working on a different task, subjects were trained to disconfirm, confirm or use no strategy on Wason's 2–4-6 task. Disconfirmatory subjects were again significantly more successful. The overall results suggest that a disconfirmatory strategy is superior for both groups and individuals on tasks that simulate scientific problems. Reasons why this effect was obtained in the present research but not in earlier studies are discussed.