The predictive validity of subjects' self-reports of the severity of four groups of symptoms associated with the premenstrual syndrome (PMS) was assessed by canonical correlation of retrospective self-reports of usual symptom severities with prospectively obtained symptom severity scores from the next two cycles. Prospective scores from the second cycle were then correlated with retrospective recall scores obtained after the end of that cycle. A measure of inter-cycle variability was obtained by correlation between two consecutive sets of prospective scores. The symptoms studied were tension, depression, cognitive and physical ('water retention') symptoms. It was found that subjects' recall of a particular cycle predicted 72% of the variance in that cycle's prospective severity scores, indicating that the subjects correctly interpreted the severity of premenstrual symptoms and distinguished them from symptoms present in the follicular part of the cycle. Retrospective reports of usual PMS symptomatology predicted 21% of the variance in symptom scores in the next menstrual cycle and 12% of the variance in the following one. Despite this decrease, averaging the scores from the two prospective cycles improved the prediction to 23%. Prospective scores from one cycle predicted only 14% of the variance in prospective scores from the next, suggesting a high degree of inter-cycle variability. Women's self-reports of their usual PMS symptomatology reflect their experience more accurately than has been thought. The finding of marked inter-cycle variability suggests that arguments for the use of a single cycle of prospective data in PMS evaluation are fallacious and that retrospective self-report may be clinically useful and relatively valid.