The objective of this study was to determine whether participation in a readers' test of a medical journal improves knowledge recall. A quasi-experimental design was used. The study population consisted of general practitioners who were divided into three groups: group A--subscribers to journal and readers' test, group B--subscribers to journal only, and group C--nonsubscribers. All physicians were interviewed over the telephone in January and June 1994 and subjected to a knowledge test that contained nine true/false questions. Interviewers were blinded to the correct responses; physicians were unaware of the study's objective and of the second interview. In April 1994, the readers' test required its participants to indicate the correctness of statements referring to information published 2 months earlier and containing the true responses to the study's knowledge test. Statistical analysis was done via an 'intention to treat' comparison of postintervention scores (minimum: 0, maximum: 9) and comparison of the change in scores from pre- to postintervention, using a two-sample t-test and paired t-test, respectively. Of 373 physicians, 24 (54%), 42 (38%), and 26 (11%) in groups A, B, and C agreed to participate. Mean scores were 5.4 +/- 0.68, 5.1 +/- 0.45, and 4.2 +/- 0.66 at preintervention and 6.8 +/- 0.77, 4.6 +/- 0.49, and 5.1 +/- 0.66 at postintervention for groups A, B, and C. Groups A and C had a statistically significant improvement in scores, with a mean change of 1.38 +/- 0.76 (p < .0002) and 0.84 +/- 0.67 (p < .02). This is the first study examining the effectiveness of readers' tests in the context of continuing medical education. Physicians participating in the readers' test had a significant improvement in knowledge scores compared to nonparticipants. [ABSTRACT FROM AUTHOR]