OBJECTIVE To determine what factors influence the precision of anticoagulant control using warfarin by examining the computerised records of 2207 patients. METHODS Records from seven district general hospitals were combined and analysed. The precision of anticoagulant control was taken as the absolute deviation of International Normalised Ratio (INR) from target at the most recent determination. This quantity was examined using univariate and multiple regression analyses. RESULTS Deviation of INR from target was continuously distributed, almost symmetrically about a mean of zero. The patients' age and sex had little bearing on control. Patients with a high target INR were more likely to be undertreated, and patients taking higher doses of warfarin were more likely to be overtreated. Previous over- or undertreatment were strongly related to poorer current control. The control of treatment varied substantially among the seven hospitals. One possible cause of this variation was the dose adjustment coefficient: the greater the dose adjustment for a given deviation from target INR, the better was the control achieved. CONCLUSIONS Several groups of patients were identified whose control was less satisfactory and in whom anticoagulant treatment needs particular scrutiny: these include patients with a record of previous over- or undertreatment, but not elderly patients in general. The variation in control among hospitals is a source of concern that merits further attention to achieve better uniformity of anticoagulant treatment.