Two studies were conducted to examine gymnasts' memory capabilities in relation to skill level (high and low) and age level (6±8 years and 14±16 years). The ®rst experiment consisted of two trials performed 7 days apart. In the ®rst trial, gymnasts were brie¯y exposed to a twelveelement ¯oor routine, and were then required to perform the routine in the same serial order. One week later they were again brie¯y presented the same sequence, but with the ®fth and seventh elements missing. The task was to perform the sequence in order, including the missing elements. The second study involved three trials that were conducted on the same day. A tenelement sequence was brie¯y presented on the ®rst trial followed by an immediate test of that sequence. One hour later, participants were asked to perform the sequence again (delayed test). An hour following that, the sequence was again presented with two elements missing. The sequence had to be performed with the missing elements. The results indicate that serial position accuracy of a gymnastic ¯oor routine improves with both age and experience, and that the older experienced gymnasts outperformed all the other gymnasts on all phases of the experiments. These results are interpreted within the Ericsson and Kintsch (1995) theory of Long-term Working Memory.