Amorphous polymers below their glass-transition temperature are inherently not at equilibrium. As a result, their structures continuously relax in an attempt to reach the equilibrium state. The current models of structural recovery can quantitatively describe the process. One of the parameters needed for the models is the nonlinearity parameter x. It has been proposed that x can be obtained from experimental data with the so-called peak-shift method. In this work, we use the Tool-Narayanaswamy-Moynihan model to identify the factors that determine the accuracy of the peak-shift method and to quantify the errors in the value of x obtained from the peak-shift method. In addition, we determine the influence of the error in x on the evaluation of the nonexponential model parameter beta. Finally, the peak-shift method is compared with the traditional curve-fitting method for model parameter determination. (c) 2002 wiley Periodicals, Inc.