Long-term semantic memory (LTM) is known for affecting recall during working memory (WM) tasks. However, the way LTM intervenes in WM remains unknown. Moreover, the available findings are incongruent concerning how attention modulates the impact of LTM on WM. To examine this issue, the involvement of LTM representations in a complex span task was manipulated through variations of the associative relatedness of memory items, while the attentional demand of the concurrent task was varied. Children and young adults were also compared, because children are less efficient in using refreshing for maintenance than adults. Despite the impact of the three main factors on recall performance, which was better for related than unrelated words, with the low rather than the high demanding concurrent task and for adults than children, there was no interaction between associative relatedness and attentional demand, neither in children nor in adults. We replicated these results in a second experiment with a more attention-demanding concurrent task. Moreover, analyses of recall latency showed that adults were faster than children at recalling words and both age groups were faster for related (vs. unrelated) words, but there was no effect of the concurrent attentional demand on recall latency and no interaction. Finally, errors were mostly omissions and transpositions, both more prevalent under high concurrent attentional demand. The present findings suggest that the availability of attention does not modulate the effect of LTM on WM. We discuss how WM models can account for this finding and how LTM can act on WM functioning.