Objective: To determine the prevalence, nature and predictors of prescribing errors (PEs) in three mental health hospitals.Setting: Inpatient units in three National Health Service (NHS) mental health hospitals in the North West of England.Participants: Trained clinical pharmacists prospectively recorded the number of PEs in newly written or omitted prescription items screened during their routine work on 10 data collection days. A multidisciplinary panel reviewed PE data using established methods to confirm (1) the presence of a PE, (2) the type of PE and (3) whether errors were clinically relevant and likely to cause harm.Primary outcome measures: Frequency, nature and predictors of PEs.Results: Of 4427 screened prescription items, 281 were found to have one or more PEs (error rate 6.3% (95% CI 5.6 to 7.1%)). Multivariate analysis revealed that specialty trainees (OR 1.23 (1.01 to 1.51)) and staff grade psychiatrists (OR 1.50 (1.05 to 2.13)) were more likely to make PEs when compared to foundation year (FY) one doctors, and that specialty trainees and consultant psychiatrists were twice as likely to make clinically relevant PEs (OR 2.61 (2.11 to 3.22) and 2.03 (1.66 to 2.50), respectively) compared to FY one staff. Prescription items screened during the prescription chart rewrite (OR 0.52 (0.33 to 0.82)) or at discharge (OR 0.87 (0.79 to 0.97)) were less likely to be associated with PEs than items assessed during inpatient stay, although they were more likely to be associated with clinically relevant PEs (OR 2.27 (1.72 to 2.99) and 4.23 (3.68 to 4.87), respectively). Prescription items screened at hospital admission were five times more likely (OR 5.39 (2.72 to 10.69)) to be associated with clinically relevant errors than those screened during patient stay.Conclusions: PEs may be more common in mental health hospitals than previously reported and important targets to minimise these errors have been identified.