Cognitive strategies are essential for visuomotor adaptation. Task set size (the number of targets) often determines which strategy is engaged: working memory (WM) is sufficient for small set size, but its capacity is often limited for more complex tasks. These WM limitations may be bypassed by retrieving a strategy from  long-term motor memory (LTM), but the required repetition is unknown. To address this, participants were trained on a single target–rotation pair, establishing an LTM visuomotor solution, after which we probed LTM-based retrieval across set sizes by comparing it with WM. The results indicated that LTM-based retrieval was relatively automatic and largely insensitive to set size, whereas WM-based retrieval declined as set size increased. Initial practice of just 10 or 30 trials produced similarly robust LTM for a strategy, which outperformed strategies held in WM. Model-based analyses suggested a trend toward higher LTM precision at larger set sizes, consistent with dynamic interactions between WM and LTM.