S U M M A R YPrecise and accurate earthquake hypocentres are critical for various fields, such as the study of tectonic process and seismic-hazard assessment. Double-difference relocation methods are widely used and can dramatically improve the precision of event relative locations. In areas of sparse seismic network coverage, however, a significant trade-off exists between focal depth, epicentral location and the origin time. Regional depth-phase modelling (RDPM) is suitable for sparse networks and can provide focal-depth information that is relatively insensitive to uncertainties in epicentral location and independent of errors in the origin time. Here, we propose a hybrid method in which focal depth is determined using RDPM and then treated as a fixed parameter in subsequent double-difference calculations, thus reducing the size of the system of equations and increasing the precision of the hypocentral solutions. Based on examples using small earthquakes from eastern Canada and southwestern USA, we show that the application of this technique yields solutions that appear to be more robust and accurate than those obtained by standard double-difference relocation method alone.Precise and accurate earthquake hypocentres are important for the study of tectonic processes, earthquake recurrence and event discriminations. In stable continental regions, this information plays a particularly important role in relating seismicity to geological structures, which can be significant for hazard studies related to critical facilities such as nuclear plants, hydroelectric dams and long-term repositories for hazardous waste. The accuracy of absolute hypocentre locations is controlled by various factors, including seismograph network geometry, the discernibility of significant phases, arrival-time reading accuracy and knowledge of crustal-velocity structure (Pavlis 1986;Gomberg et al. 1990). With the exception of areas with a dense distribution of seismograph stations (e.g. Japan and California), the location uncertainty of routinely determined hypocentres is typically many times larger than the source dimension of the events (Waldhauser & Ellsworth 2000), rendering them ill-suited for fine-scale correlation with geological structures.Various techniques have been developed to improve the precision and accuracy of earthquake hypocentres. Dodge et al. (1995) converted cross-correlation delay times to absolute times by assigning them to ordinary time picks. Using a joint hypocentre determination method, the traveltimes are then inverted to simultaneously estimate hypocentral parameters, velocity model corrections and station corrections. Waldhauser & Ellsworth (2000) presented a double-difference technique, known as hypoDD. Their method allows the simultaneous relocation of large numbers of earthquakes over large distances. This method determines distances between correlated events that form a single multiplet, limited by the uncertainty (width) of the cross-correlation peak, while simultaneously determining the relative locations ...