PurposeMedication errors (MEs) are important in terms of their magnitude and severity, and there are numerous systems in place to reduce their occurrence. However, the ideal system has not yet been identified. The authors' institution uses three different medication prescription‐dispensation systems which operate simultaneously. ME rates were compared, overall and by phase (prescription, transcription and administration) and their overall and specific clinical impact.Design/methodology/approachThe administration of medicinal products was observed directly and compared with medical and nursing prescriptions. Errors and adverse events were classified by a consensus of experts.FindingsIn the traditional system the error prevalence rate was 13.59 per cent, (99 per cent CI, 12.15‐14.61 per cent), in the single dose system it was 6.43 per cent (99 per cent CI, 5.53‐7.32) and in the electronic prescription system it was 8.86 per cent (99 per cent CI, 7.33‐10.17). The highest error rates in all phases were found in the traditional system. The phase affected by most errors in all three models was transcription, and the least affected was administration, except for the single dose system, in which prescription was the worst. The effects of errors in the administration phase are greater, although less so than with the automated system.Research limitations/implicationsThe dispensation phase was not analyzed.Practical implicationsA study of errors will enable us to reduce their occurrence if we know the most frequent types and in which phase they are produced, we will be able to prioritise the areas in which to work and select the necessary preventive measures.Originality/valueIt is possible that automated medication dispensation systems reduce error rates and the severity of their effects.