The sense of controlling our own actions, and through them events in the outside world, is a fundamental feature of human mental life. The mechanisms underlying this “sense of agency” are poorly understood, though animal learning studies have revealed the basic brain mechanisms of learning and prediction on which sense agency seems to depend. We recently showed (Wenke, Fleming and Haggard) that sense of agency depends not only on the actual result of our actions, but on how efficiently we select which action we perform. When actions were subliminally primed with a compatible left/right cue, participants felt greater control over a subsequent visual effect triggered by the action, compared to when the primes were incompatible. These primes did not predict the effects of action, but simply influenced the ease of action selection. Thus, this aspect of the conscious experience of agency must be prospective, rather than being retrospectively based on action-outcome relations. People frequently report the experience of knowing exactly what to do in a given situation: we suggest this reflects a sense of agency based on selecting what to do, rather than based on action outcomes. WP6 will study the role of instrumental learning in conscious sense of agency. We hypothesise that people must first learn the relation between actions and their consequences: only then can efficiency of action selection be a reliable component of sense of agency. That is, action selection can only contribute prospectively to sense of agency if participants have learned how the selected action actually influences external outcomes. We therefore predict that the contribution of subliminal priming of actions to conscious sense of agency should develop during action-effect learning. To test this hypothesis, participants will learn a statistical relation between 4 different response keys and 4 different shapes of a visual stimulus caused by the keypress. For example, response key 1 will produce a square on 80% of trials, but a circle, triangle or diamond on the remaining 20%. Response 2 will produce a circle on 80% of trials, but a square, triangle or diamond on the remaining 20% and so on.