BACKGROUND: Existing criteria for the diagnosis of headache do not indicate which clinical features are most important to establish or exclude primary headache diagnoses. OBJECTIVE: To determine the value of history taking in the evaluation of patients with primary headache syndromes. METHODS: A MEDLINE search of English-language studies published since 1966 and a hand search of bibliographies from retrieved articles. Included studies cited the sensitivity of clinical and historical features for the diagnosis of migraine, tension-type, and cluster headaches. Studies were classified according to their use of the International Headache Society criteria or other criteria as the reference standard for diagnosis. RESULTS: The features most predictive of migraine, when compared with tension-type headache, are nausea, photophobia, phonophobia, and exacerbation by physical activity. Positive likelihood ratios (95% confidence intervals) are 19.2 (15.0-24.5), 5.8 (5.1-6.6), 5.2 (4.5-5.9), and 3.7 (3.4-4.0), respectively. Headache precipitants are present equally in patients with migraine and tension-type headache, with the exception of food triggers, which confer a positive likelihood ratio of 3.6 (2.8-4.6) for the diagnosis of migraine. Visual aurae are present in 84% of patients with migraine with aura. Cluster headache is strictly unilateral, is periorbital in 80% of patients, and lasts less than 1 hour in 54% of patients. CONCLUSIONS: The features most predictive of the diagnosis of migraine, when compared with tension-type headache, are nausea, photophobia, phonophobia, exacerbation by physical activity, and an aura. Cluster headache is a distinct clinical syndrome.