Flexibly switching between tasks is crucial for goal-directed adaptive behavior. Task switches require inhibition of the no-longer relevant and updating of the newly relevant task set, resulting in lower performance on trials requiring a task switch vs. repetition (i.e., switch costs). While children show greater switch costs than adults, theorized to be partly due to less distinct task-set representations, age differences can be mitigated by training. Studies in adults suggest that neural task-set representations become more distinct with training, enabling more effective processing in the lateral prefrontal cortex (lPFC). However, it is still unclear how changes in neural activation and representations contribute to training-related improvements of task switching in children. We addressed this question using univariate and multivariate analyses of fMRI data in children (8–11 years) who completed nine weeks of task-switching (N=32) or single-task (N=30) training. With training, the task-switching training group showed improved performance, accompanied by decreased activation in lPFC during task switching. Neural task-set representations in the lPFC were less distinct on switch compared to repeat trials prior to training and did not change with training in either training group. Switch-related decreases in decoding accuracy were comparable between children and a group of adults (20–30 years, N=53) who completed a single fMRI session, with no training. These results do not support the prediction that task-switch training in childhood increases distinctiveness of the specific task-set representations, but suggest general improvements in cognitive control processes, as evident in reduced lPFC activation supporting the management of relevant rules.