A self-administered questionnaire, the Health Survey Questionnaire (HSQ) was distributed to patients registered with 47 group general practices. The HSQ assesses alcohol consumption using a quantity frequency scale and includes the four CAGE questions and a question on whether respondents think they have an alcohol problem. A random stratified sample of those patients who returned an HSQ (2666 men and 1537 women) were subsequently invited to attend their general practice for an interview with the practice nurse, where weekly alcohol consumption was estimated using both a quantity frequency scale and a systematic enquiry about alcohol consumption for the week immediately preceding the interview. The latter method was taken as the 'gold standard'. Excessive drinkers were defined as men whose weekly consumption by this method was not less than 35 units per week and women drinking at least 21 units per week. After weighing the results to take account of the sampling bias in favour of the excessive drinkers, 11.7% of men and 2.9% of women were excessive drinkers according to the estimate of alcohol consumption at interview. This compares with the 7.6% of men and 2.7% of women who were heavy drinkers by the HSQ quantity frequency scale. The two interview estimates were comparable but in general the HSQ tended to underestimate consumption compared with these estimates. The questionnaire was found to be most effective in screening for excessive drinkers if all the patients who indicated concern about their drinking (i.e. those who were with CAGE positive or had a self-assessed drinking problem), as well as all of those who were above the limits for the trial on the quantity frequency scale, were selected as being potentially excessive drinkers. In the weighted sample, 14.8% of men and 6.9% of women were in this group. Using these selection criteria and taking the interview as the standard for determining the excessive drinkers a sensitivity of 58.6% and specificity of 91.08% was obtained for men, with a positive predictive value of 46.1% and a negative predictive value of 94.3%. For women the test was more sensitive (69.7%) and more specific (95.0%) and had a better negative predictive value (99.1%) than for men. The positive predictive value for women at 29.6% was not as good as that obtained for men. This analysis shows that the HSQ is a fairly effective tool for detecting excessive drinkers in general practices with a small proportion of false positive results. It is both economical and acceptable to patients in a wide range of practice settings.