Objective. - To test by means of ROC analysis the performance of the Symptom Questionnaire (SQ) of Kellner and Fava against an interview designed for making DSM-III diagnoses in 517 subjects. To analyse the trend of SQ scores at 3 months, 1 year and 2 years after the first interview in a sample of 112 persons. Design. - Cross-sectional and longitudinal analysis of the distributions of SQ scores in DSM-III cases and in non-cases. Setting. - General practice, hospital medical wards and emergency departments of Regione Emilia Romagna. Main outcome measures. - Symptom Questionnaire, Bech Mini-compendium to formulate DSM-III diagnoses. Results. - The SQ reached high sensitivity and specificity levels (80%, 76% in general practice, 86%, 74% in hospital medical wards, 83%, 85% in emergency departments with the cut-off scores of 24, 24 and 26, respectively). SQ scores show a similar trend in cases and in non-cases, with a significant drop at three months. Conclusions. - The SQ seems to be an adequate screening tool in the above-mentioned settings, but its length and the assistance sometimes needed during the compilation make its use in the emergency departments problematic.