The concept of severe migraine was raised to define migraineurs most in need of care and for use in clinical practice. We aimed to measure the frequency of severe attacks in a working sample of 276 migraine sufferers using a diary over a 3-month period. Migraine sufferers recorded each attack's clinical features, the degree of their disability, their use of drugs and the effectiveness of the drugs. Since the definition of severe attack is not standardized, we studied the impact of different definitions on the frequency. The frequency of severe attacks was 0.9% and appeared to be very sensitive to the definitions, ranging between 0.4 and 13%. In France, the extrapolated number of severe attacks is nearly one million out of a total of 115 million. In the migraineurs who had had at least one severe attack, the individual variability of intensity, duration or disability was very high, so the proportion of severe attacks in a given sufferer was low--between 15% and 50%. We conclude that the global concept of severe migraine is not relevant and should be split into two components--severe attack and severe migraine sufferer. The goals are different, too. Regarding treatment, for example, the severe attack concept is more valid for acute treatment strategies, whereas the severe migraine sufferer concept should be preferred to determine the need for prophylactic treatment. Since much work is being done nowadays to define acute treatment strategies, definition of the criteria of severe attack and validation of a measurement tool should be a priority.