Artificial muscles made be twisting and coiling polymer fibers provide outstanding performance. However, these materials show inconsistency in their non-loaded length that depends on their thermo-mechanical history. Typically, this behavior has been treated by ���training��� the samples before any actuation testing. A change in sample length occurs during training but remains consistent during subsequent heat/cool cycles at the same applied load. In this study, the training effect is investigated for a twisted and coiled nylon yarn heated over two temperature ranges: 25���50��C and 50���75��C. The training effect was most obvious in the lower temperature range, but nearly absent in the higher temperature range. When loaded below the glass transition temperature (Tg ~ 40��C) the viscoelastic strain occurs slowly but is rapidly released when the sample is first heated above Tg. The net effect of the first heating through Tg after loading is a small length change because the contraction due to actuation is offset by the expansion due to the release of the viscoelastic strain. A simple spring and dashpot model was developed and by changing only two relaxation times it was possible to simulate the observed training phenomena.