Aims: To compare baseline characteristics of clients initially preferring abstinence with those preferring non-abstinence at the screening stage of a randomized controlled trial of treatment for alcohol problems (UKATT) and to identify predictors of goal preference from client characteristics present before the preference was stated. Methods: From discussions with clients entering the trial (N = 742), screeners noted whether clients were aiming for abstinence ‘probably yes’ or ‘probably no’. Differences between the two groups thus formed were explored by univariate comparisons among client characteristics recorded at baseline assessment and by logistic regression analysis with pre-existing characteristics as independent variables. Results: Across all UKATT sites, 54.3% of clients expressed a preference for abstinence and 45.7% for non-abstinence. In univariate comparisons, clients preferring abstinence were significantly (P <0.01) more likely to: (i) be female, (ii) be unemployed, (iii) report drinking more heavily but less frequently, (iv) have been detoxified in the 2 weeks prior to assessment, (v) report more alcohol problems, (vi) be in the action stage of change, (vii) report greater negative expectancies of drinking, (viii) report greater mental and physical ill-health, (ix) report less social support for drinking and (x) be more confident of their ability to resist heavy drinking in tempting situations. In the logistic regression model, the strongest predictors of goal preference were gender, drinking pattern, recent detoxification and social support for drinking. Conclusion: The implications of these findings for service delivery are best considered in conjunction with findings from a companion paper reporting treatment outcomes associated with each goal preference.