Earthquake source time functions carry information about the complexity of seismic rupture.We explore databases of earthquake source time functions and find that they are composed of distinct peaks that we call subevents. We observe that earthquake complexity, as represented by the number of subevents, grows with earthquake magnitude. Patterns in rupture complexity arise from a scaling between subevent moment and main event moment. These results can be explained by simple 2-D dynamic rupture simulations with self-affine heterogeneity in fault prestress. Applying this to early magnitude estimates, we show that the main event magnitude can be estimated after observing only the first few subevents.Plain Language Summary Seismograms are measurements of waves from earthquakes.They give us information about what happened on the fault at the place where the earthquake occurred. Seismograms can be difficult to interpret because they are often very complicated. Why? One reason is that the waves change when they travel long distances between the fault and a seismometer. Seismologists correct for this effect, however, by constructing something called a source time function. Source time functions are much easier to understand than raw seismograms. We examine a catalog of source time functions from around the world. We find that large earthquakes are composed of many smaller events that we call subevents. The size of a subevent is related to the size of the main earthquake. One important outcome is that we can predict the final size of an earthquake after observing only the first few subevents.