Some people who have had a brain injury can be verbally or physically aggressive, and require rehabilitation to help reduce these behaviours and assist them in their recovery. Mental health professionals need to accurately measure the level and type of aggression so that they can judge treatment effectiveness. There are different methods of measuring aggression. One way is to record both the frequency and actual severity of observed incidents. In brain injury settings the 'Overt Aggression Scale - Modified for Neurorehabilitation' (OAS-MNR) was developed to do this. Some have argued that the results of aggression do not always fully reflect the seriousness of the incident. For example, if a nurse was attacked with a knife but she managed to disarm the aggressor then the actual outcome might not be serious, but the potential severity was very high. Another way of measuring aggression, therefore, is to assess the potential severity of the incident and how much intent the aggressor had to inflict damage. The 'Attempted and Actual Assaults Scale' (Attacks) has been developed to do this, but it has not previously been used in a brain injury setting. This study used both the OAS-MNR and Attacks scales to measure aggression on a brain injury unit. We found that information in the Attacks scale about potential severity and intent to do harm added validity to the observed behaviour recorded on the OAS-MNR. It may be helpful in future to routinely capture information about potential severity by modifying the OAS-MNR. Aggression is a consequence of acquired brain injury that may necessitate admission to neurobehavioural services. The 'Overt Aggression Scale - Modified for Neurorehabilitation' (OAS-MNR) is a valid, reliable means of capturing this. A criticism of observational rating scales is they do not reflect factors like intent to harm which results in recording anomalies. 'Attacks' has been proposed as a measure which achieves this within psychiatric settings. Principal goals of this study are to determine the usefulness of measuring similar concepts in neurobehavioural services and further validating both scales. A total of 1066 physical assaults were recorded in 6 weeks by 25 patients in an inpatient neurobehavioural programme using the OAS-MNR. Fifty incidents were also rated on Attacks. Convergent validity for using both measures in neurobehavioural services was found. Modifying OAS-MNR severity scores using one of two factors found to underlie Attacks produced an index that successfully discriminated incidents whose risk necessitated more intrusive intervention, which was not evident otherwise. Modifying scores that objectively reflect severity of physical assaults using measures of perceived intent should be a feature of observational recording scales such as the OAS-MNR. Ensuring robust inter-rater reliability will be essential in any development work.