We combine geological and seismological data to describe the shape of the magnitude–frequency distribution for the Canterbury earthquake sequence. In doing so, we take the opportunity of the new datasets to address a long‐standing debate in the literature as to whether the seismicity of individual faults or fault zones is best described by the Gutenberg–Richter (GR) relationship or the characteristic earthquake (CE) model. We find that the magnitude–frequency distribution for the entire composite fault zone of the Canterbury earthquake sequence is adequately described by the GR relationship when uncertainties in the GR curve (fitted to the instrumental catalog data) and the range of paleoseismically derived recurrence intervals for large (mainshock) earthquakes on the Greendale fault are collectively considered. In contrast, the magnitude–frequency distribution for the smaller area of the Greendale fault is better described by the CE model. The difference is one of scale, in that the composite fault zone represented by the Canterbury earthquake sequence shows a GR distribution, whereas individual faults within the zone show CE distributions. Definition of magnitude–frequency distributions for seismic‐hazard modeling must therefore take scaling considerations into account, rather than simply assuming that a single magnitude–frequency distribution shape is applicable to all fault definitions.