Task-cuing paradigms are typically taken to explore control of task-set. However, they can be construed as requiring not selection of a task-set, just retrieval of a cue+stimulus-->response (CSR) mapping. In this paper we considered performance in a task-cuing paradigm in which participants saw a color cue that indicated whether they should classify a digit as odd/even or high/low using one of two responses. Half the participants were instructed in terms of tasks (Task group) whilst the others were required to learn the CSR mappings without mention of tasks (CSR group). Predicted performance under CSR conditions was modeled using an APECS connectionist network. Both the model and CSR group produced small switch costs, mostly due to incongruent stimuli, and large congruency effects that reduced with practice. In contrast, the Task group produced a larger switch-cost and a smaller, stable congruency effect.