Recent studies using Stroop's paradigm have shown that word recognition processes can be controlled when the local context of the task is manipulated. In the present study, factors related to the participants' broader context (i.e., presence vs. absence of a competitor and of a desired reward) were manipulated. The results (1) support the conclusion that control of semantic-level activation can be unconscious but effective versus conscious but ineffective, (2) suggest that unconscious control alone operates on line (i.e., when the participant is responding), and (3) clarify the impact of socio-contextual factors that have been confounded in past research. Taken together, these findings strengthen the view that word recognition processes are controllable and offer new reasons to pay constant attention to the social environment of cognition.