The Antarctic marginal ice zone, the regularly wave‐affected outer band of the sea ice covered Southern Ocean, typically contains an unconsolidated ice cover comprised of smaller, thinner floes than the inner ice pack. Thus, it is a highly dynamic region and susceptible to rapid expansion and contraction, making it a focal area for understanding and predicting the response of Antarctic sea ice to a changing climate. This novel study uses unsupervised statistical clustering of sea ice data simulated by a global sea ice model (standalone CICE6 combined with a wave propagation module and prescribed ocean) to address the outstanding challenge of separating the marginal ice zone from the inner ice pack in sea ice data sets. The method identifies a marginal ice zone with the desired characteristics and floe size is shown to be the key variable in the classification. Simulated marginal ice zone widths are similar to those derived from satellite observations of wave penetration distances, but contrast with those using the standard 15%–80% areal sea ice concentration proxy, particularly during austral winter. The simulated marginal ice zone is found to undergo a seasonal transition due to new ice formation in winter, increased drift in spring, and increased rates of wave‐induced breakup and melting in summer. The understanding gained from the study motivates incorporation of wave and floe‐scale processes in sea ice models, and the methods are available for application to outputs from high‐resolution and coupled sea ice–ocean–wave models for more detailed studies of the marginal ice zone (in both hemispheres).