The subject of this study is non-response to mail questionnaires and supplementary telephone interviews in evaluation of outcome of treatment for alcohol dependence. The study is based on two clinical populations (n= 121 and 603) from Runnagården, Sweden, a coercive care setting with both voluntary and compulsorily committed patients. The populations differed regarding size, treatment and background factors. Replies were obtained from two- and one-third of the patients in the two populations respectively. Replies from social workers were obtained in 67% and 57% of the cases respectively, and from significant others in 59% of a smaller sub-sample of the second population. With respect to the combination of patient and collaterals, we received at least one questionnaire reply concerning 96% and 79% of the two populations respectively. In this study we analysed non-response and evaluated two methods to remedy this, i.e., prediction of the outcome in non-responders from additional telephone follow-up and questionnaires to collaterals. About two-thirds of patients who replied had improved, compared with about one-third of non-responders. Significant others replied more often if patients had improved, while social workers replied more often if patients had not improved. Patients who did not reply to mail questionnaires, but answered the same questions in a telephone interview, differed in drinking outcome from non-responders, but not from mail respondents. The method of using questionnaires addressed to social workers as an additional source of data can be used in countries with a general system of social welfare services and is better than the use of significant others as collaterals. In conclusion, replies obtained by additional telephone interviews were not representative for patients who did not return mail questionnaires. Non-response groups were highly selected in each of the three questionnaires (addressed to patients, significant others and social workers). Since the selectivity had opposite tendencies, the combination of questionnaires to patients and to social workers seems to be the most representative.