Response-selection bottleneck (RSB) models provide an important account of dual-task interference. These models are thought to have difficulty, however, explaining the commonly observed backward compatibility effect (BCE), which is the finding that the speed of first-task responses depends on their compatibility with later second-task responses. One explanation of this phenomenon is based on a distinction between an early response activation stage and a subsequent response selection stage (e.g., Hommel, 1998). The former stage is thought to be influenced by both tasks and to be responsible for BCEs, whereas the latter stage implements the RSB suggested to explain many aspects of dual-task interference. The present experiments measured lateralized readiness potentials (LRPs) to test predictions of this early response activation account. The results are generally consistent with a distinction between response selection and response activation, but they suggest that BCEs arise because second-task stimuli influence first-task response selection rather than because these stimuli activate the second-task responses with which they are associated. Thus, the results support an account of BCEs that is even more consistent with RSB models than previous accounts.