OBJECTIVE: To determine the accuracy of clinicians' visual observations of gait disorders following traumatic brain injury (TBI). METHODS: 30 ambulant participants (sample of convenience) receiving physiotherapy for mobility limitations following TBI and 25 age, height, weight and sex matched healthy unimpaired controls (HC) were recruited. Kinematic and ground reaction force data during gait were captured and video recordings were concurrently collected. Participants with TBI walked at self-selected speed whilst HCs walked at preferred speed as well as the mean TBI speed for comparison. 40 doctors, experienced physiotherapists, new graduate physiotherapists and novices were observers. Each viewed and rated 36 gait variables for a randomized sub-sample of 10 participants with TBI. Observer inaccuracy was calculated for each gait variable. RESULTS: Overall the accuracy of observational gait analysis was low and there was considerable variability in observations between clinicians. For most kinematic variables, observer inaccuracy ranged from 30% to 50%. Although experienced observers were generally more accurate, average inter-item correlations were low, indicating that experience did not consistently improve the accuracy of visual observations. Observational plane, gait variable type, the joint or the segment had little effect on accuracy of observations. CONCLUSIONS: Observational gait analysis for adults with TBI has relatively low accuracy. Some of the gait abnormalities evident from quantitative gait analysis were not detected by observational gait analysis.