Objective: To investigate the comparative 2-year efficacy of brief alcohol interventions delivered in-person versus through computer-generated feedback letters among general hospital inpatients with at-risk alcohol use. Methods: In 2011–2012, all general hospital inpatients aged 18–64 years on 13 wards at 4 medical departments in 1 general hospital were systematically screened for at-risk alcohol use. Nine-hundred sixty-one inpatients who screened positive for at-risk alcohol use and negative for more severe alcohol problems were randomized by timeframe to in-person counseling (PE), computer-generated feedback letters (CO) and assessment only (AO). Both interventions included 3 contacts: on the ward and 1 and 3 months later and were based on the transtheoretical model (TTM). After 6, 12, 18, and 24 months computer-assisted telephone interviews assessed self-report outcomes: gram alcohol per week (primary), at-risk alcohol use and highest blood alcohol concentration (both secondary). Latent growth models were used. Results: After 24 months, CO resulted in a greater reduction of self-reported gram alcohol per week than AO (p = .027); PE did not differ significantly from AO (p = .503) and CO (p = .088); and group differences concerning secondary outcomes were not statistically significant (ps > 0.07). After 6 months, the odds of at-risk alcohol use were reduced by half in PE versus AO (odds ratio = 0.50, 95% confidence interval: 0.25–0.98). Conclusions: No significant group differences between in-person counseling and computer-generated feedback letters were found. TTM-based computer-generated feedback letters reduced self-reported gram alcohol over 2 years and can be considered a long-term effective alternative in medical settings when addressing at-risk alcohol use.