Some 67 journal articles that described and evaluated health education programmes in developing countries were read by two independent reviewers who examined the methodology used in the studies. Of the articles 47% provided a sufficiently detailed description of the educational intervention to allow replication and 40% described the educational level of the intended audience. Only 21% were controlled studies employing sample sizes greater than 60 individuals or two clusters, although six studies used randomized or quasi-randomized designs. Of the studies 33% looked at changes in health status while another 33%used observable changes in health behaviour as an endpoint. There was good agreement between the reviewers on whether these characteristics were present. Only three of the articles contained all four methodological attributes described above. The results of these articles suggests that successful health education depends on using a few messages, of proven benefit, repeatedly, and in many forums. It is important to improve the methodological quality of health education research. This can be done by using controlled, preferably randomized, designs, ensuring adequate sample sizes, examining only objective changes in behaviour or, better yet, changes in morbidity or mortality. Research reports should describe in detail the educational intervention employed and the target audience.