An approach to analysis of time series of edge localized modes (ELMs) is proposed. It is based on the use of the autoregressive moving average model, which decomposes time series into deterministic and noise components. Despite the inclusion of nonlinearity in the model, the resulting deterministic equations for the ELM time series measured on Axially Symmetric Divertor Experiment Upgrade tokamak turn out to be linear. This contrasts with the findings on JAERI tokamak (JT-60U) and tokamak a configuration variable that ELMs exhibit features of chaotic dynamics, namely, the presence of unstable periodic orbits. This methodology for distinguishing chaotic behavior is examined, and found to be susceptible to misinterpretation. (C) 2004 American Institute of Physics.