The objective was to assess the incidence and consequences of medication errors, highlight sources of recurrent error and institute changes in practice to prevent their recurrence. Utilising a continuous quality improvement approach, a 2-year prospective cohort study was undertaken using an adverse incident reporting scheme. A multidisciplinary committee analysed medication error reports, classifying them according to type (prescription, supply or administration), severity (serious or not serious) and clinical outcome. Changes in policy and practice were implemented to reduce the frequency of errors. There were 441 reported medication errors in the study period, during which 682 patients were admitted for 5315 inpatient days. Errors were more seven times likely to occur in the intensive care setting. Doctors accounted for 72% of errors and prescription errors doubled when new doctors joined the rotation. Most errors (68%) were detected prior to drug administration. Twenty-four serious medication errors were not detected in advance, but only 4 had overt clinical consequences. Excluding prevented errors and appropriate deviations from prescribed therapy, there were 117 actual medication errors (1/5.8 admissions, or 1/45 inpatient days). During the 2nd year of the scheme, the incidence of all reported errors, administration errors and serious errors fell, but the prescription error rate remained constant.