Primary headaches are frequent in children. They are difficult to diagnose because there is much disagreement about the interpretation of the historical data and the use of signs and/or symptoms in diagnosis. It would be useful, therefore, to standardize this procedure. We used linear discriminant analysis to determine a classification rule capable of diagnosing new cases of chronic and recurrent primary headache in children. We considered 23 symptoms in 118 patients. Through discriminant analysis we chose five variables: frequency of the attacks, type of pain, neurologic deficits, nausea, and vomiting. With this classification rule, we obtained a total correct classification of 84.7% for migraine, psychogenic headache, and non-defined headache in respect to the diagnoses formulated by a pediatrician and a child neuropsychiatrist after 3 months of follow-up. Our method for diagnosing migraine has a sensitivity of 95% and a specificity of 100%. The algorithm, validated on another 105 pediatric patients, produced a total correct result of 82.9%.