Characteristic timescales for the Northern Annular Mode (NAM) and Southern Annular Mode (SAM) variability are diagnosed in historical simulations submitted to the Coupled Model Intercomparison Project Phase 5 (CMIP5) and are compared to the European Centre for Medium-Range Weather Forecasts ERA-Interim data. These timescales are calculated from geopotential height anomaly spectra using a recently developed method, where spectra are divided into low-frequency (Lorentzian) and high-frequency (exponential) parts to account for stochastic and chaotic behaviors, respectively. As found for reanalysis data, model spectra at high frequencies are consistent with low-order chaotic behavior, in contrast to an AR1 process at low frequencies. This places the characterization of the annular mode timescales in a more dynamical rather than purely stochastic context. The characteristic high-frequency timescales for the NAM and SAM derived from the model spectra at high frequencies are ∼5 days, independent of season, which is consistent with the timescales of ERA-Interim. In the low-frequency domain, however, models are slightly biased toward too long timescales, but within the error bars, a finding which is consistent with previous studies of CMIP3 models. For the SAM, low-frequency timescales in November, December, January, and February are overestimated in the models compared to ERA-Interim. In some models, the overestimation in the SAM austral summer timescale is partly due to interannual variability, which can inflate these timescales by up to ∼40% in the models but only accounts for about 5% in the ERA-Interim reanalysis.