Abstract Background Current tasks measuring social cognition are usually ‘pen and paper’ tasks, have ceiling effects and include complicated test instructions that may be difficult to understand for those with cognitive impairment. We therefore aimed to develop a set of simple, instructionless, quantitative, tasks of emotion recognition using the methodology of eye tracking, with the subsequent aim of assessing their utility in individuals with behavioural variant frontotemporal dementia (bvFTD). Methods Using the Eyelink 1000 Plus eye tracker, 18 bvFTD and 22 controls completed tasks of simple and complex emotion recognition that involved viewing four images (one target face (simple) or pair of eyes (complex) and the others non-target) followed by a target emotion word and lastly the original four images alongside the emotion word. A dwell time change score was then calculated as the main outcome measure by subtracting the percentage dwell time for the target image before the emotion word appeared away from the percentage dwell time for the target image after the emotion word appeared. All participants also underwent a standard cognitive battery and volumetric T1-weighted magnetic resonance imaging. Results Analysis using a mixed effects model showed that the average (standard deviation) mean dwell time change score in the target interest area was 35 (27)% for the control group compared with only 4 (18)% for the bvFTD group (p < 0.05) for the simple emotion recognition task, and 15 (26)% for the control group compared with only 2 (18)% for the bvFTD group (p < 0.05) for the complex emotion recognition task. Worse performance in the bvFTD group correlated with atrophy in the right ventromedial prefrontal and orbitofrontal cortices, brain regions previously implicated in social cognition. Conclusions In summary, eye tracking is a viable tool for assessing social cognition in individuals with bvFTD, being well-tolerated and able to overcome some of the problems associated with standard psychometric tasks.