Background. Relatively little is known about medication administration errors in mental health settings. Objective. To investigate the frequency and nature of medication administration errors in old-age psychiatry. To assess the acceptability of the observational technique to nurse participants. Method. Cross-sectional study technique using (i) direct observation, (ii) medication chart review and (iii) incident reports. Setting. Two elderly long-stay wards in an independent UK psychiatric hospital. Participants. Nine nurses administering medication at routine medication rounds. Main outcome measures. Frequency, type and severity of directly observed medication administration errors compared with errors detected by retrospective chart review and incident reports. Results. Using direct observation 369 errors in 1423 opportunities for errors (25.9%) were detected vs. chart review detected 148 errors and incident reports none. Most errors were of doubtful or minor severity. The pharmacist intervened on four occasions to prevent an error causing patient harm. The commonest errors observed were unauthorized tablet crushing or capsule opening (111/369, 30.1%), omission without a valid reason (100/369, 27.1%) and failure to record administration (87/369, 23.6%). Among the nurses observed, the error rate varied widely from no errors to one error in every two doses administered. Of the seven nurses who completed the post-observation questionnaire, all said they would be willing to be observed again. Conclusion. Medication administration errors are common and mostly minor. Direct observation is a useful, sensitive method for detecting medication administration errors in psychiatry and detects many more errors than chart review or incident reports. The technique appeared to be acceptable to most of the nursing staff that were observed.