Three experiments using psychological refractory period (PRP) tasks documented backward crosstalk effects in which the nature of the second-task response influenced the first-task response latencies. Such effects are difficult to explain within currently popular bottleneck models, according to which second-task response selection does not begin until first-task response selection has finished. In Experiments 1 and 2, the first of the PRP tasks required a choice reaction time (RT) response, whereas the second task required a go/no-go decision. Task 1 responses were faster when the second task required a go response than when it required a no-go response. Experiment 3 showed that Task 1 RTs were also influenced by the complexity of second-task responses. These backward crosstalk effects indicate that significant second-task processing is carried out in time to influence first-task responses and thus challenge strictly serial bottleneck models.