Treatment of migraine attacks and prophylactic treatment of migraine are each discussed under the four headings: patient selection, trial design, evaluation of results and statistics. Checklist of problems encountered in these trials are given in the tables. The unsuitability of currently used migraine definitions for scientific investigations is stressed. Operational diagnostic criteria for common and classic migraine are given. No clear separation of common migraine attacks and interval headaches is possible, but the problem can be reduced by setting an upper limit (4-6/month) on migraine attack frequency. For the treatment of migraine attacks, the crossover design should always be used. We suggest dose-response studies to solve the problem of equipotency of doses, when 2 drugs are compared. A prophylactic drug should be studied both with the crossover design and with the less powerful group comparison design. Evaluation of results should be based on patients' attack report forms and, in prophylactic studies, a headache diary. We suggest global rating of attack severity. A rather simple headache index (sum of severity scores for each day with migraine) and perhaps a sum of global scores should be used in prophylactic trials. Confidence limits and power should be given, particularly when statistically insignificant results are reported. Migraine frequency often decreases with time regardless of treatment, and this 'time effect' should be separated from the therapeutic effect by appropriate statistical methods.