OBJECTIVE Historically, drinking was described in terms of the binary state of drinking or abstinence. More recently, drinking is viewed as a dynamic process, a trajectory of multiple fluctuations between drinking and abstinence across time. Modeling that kind of dynamic requires measuring individuals' drinking longitudinally during extended periods at cadences that capture the oscillations. Retrospective timeline followback instruments, such as the Form 90, capture self-reported drinking on a daily basis. However, prior reliability estimates were based on variables averaged across cadences of 30 days. This study expands the evaluation of retest reliability to daily reports of drinking. METHODS A secondary analysis of previously published data was conducted. The analysis sample (n=70) consisted of treatment seeking adult men (80%) and women. Initial interview data were collected using the Form 90. Participants were re-interviewed 2 days later by different interviewers. A 90-day retrospective window was indexed for both interviews, resulting in a common recall period of 88 days. RESULTS Agreement was assessed using the kappa statistic and the intraclass correlation. Kappa for each day in the indexed period ranged from kappa = .481 to kappa = .971. The unadjusted average level of agreement across the recall period was kappa = .766 (95% confidence interval = .750-.782). CONCLUSIONS Study results indicate that self-report of daily drinking has excellent retest reliability. These results suggest that data collected with the Form 90 can be used to model drinking with cadences that capture daily fluctuations in drinking.