The Cone of Direct Gaze (CoD) is described as the range ofeye gaze deviations over which an observer reports gaze asbeing directed towards them. The CoD has been found tonarrow with age across childhood (Mareschal et al. 2016). Weinvestigated whether reinforcement learning, so critical inshaping eye gaze responses in infancy, was able to accountfor the emergence of a CoD and its narrowing in childhood.To this end, we adapted Triesch et al.'s (2006) reinforcementlearning model by (1) defining a topology over objectlocations, and (2) introducing opponent non-linear rewardprofiles for looking at objects and caregivers. In Simulation 1we show that these modifications give rise to a functionalCoD in which there is reduced eye gaze following andincreased fixation on the caregiver for locations with a smallcaregiver eye gaze eccentricity. In Simulation 2 we show thatthe width of this effect reduces with learning, suggesting thatdevelopmental decreases in the CoD may be driven byreinforcement learning. In Simulation 3 we explore howchanges in model parameters can explain the CoD in highanxiety populations. Finally, the model provides one way ofunifying the developmental gaze-following and CoDliteratures, until now considered largely independent.