The interpretation of a cloud of earthquake hypocenters in terms of causative structures is not a simple task. Locations are subject to uncertainties, which will not be the same for every earthquake. The data should therefore not be interpreted simply by inspection, which is difficult in the case of three-dimensional data anyway. Instead, we propose using the location uncertainties as a guide in processing the data. Earthquake locations are moved inside their uncertainty or confidence ellipsoids until a simplified picture of the earthquake cloud is obtained, which can then be interpreted in terms of some simplified structure such as faults. The aim of the approach is to give the simplest possible structure that is consistent with all the location and confidence ellipsoid data. The method is applied to three synthetic sets of data. These illustrate the potential and limitations of the method. Application to a real earthquake data set from Rabaul Caldera in Papua New Guinea gives an image of the caldera ring fault that suggests departures from the simple ring-fault structure previously assumed. Sensitivity analysis on the Rabaul data shows that the method is not unduly sensitive to the assumptions that have to be made in applying it.¿ 1997 American Geophysical Union