We simultaneously reanalyzed two decades (1984–2003) of the digital seismic archive of Northern California using waveform cross-correlation (CC) and double-difference (DD) methods to improve the resolution in hypocenter locations in the existing earthquake catalog generated at the Northern California Seismic Network (NCSN) by up to three orders of magnitude. We used a combination of ∼3 billion CC differential times measured from all correlated pairs of events that are separated by less than 5 km and ∼7 million P wave arrival-time picks listed in the NCSN bulletin. Data were inverted for precise relative locations of 311,273 events using the DD method. The relocated catalog is able to image the fine-scale structure of seismicity associated with active faults and revealed characteristic spatiotemporal structures such as streaks and repeating earthquakes. We found that 90% of the earthquakes have correlated P wave and S wave trains at common stations and that 12% are colocated repeating events. An analysis of the repeating events indicates that uncertainties at the 95% confidence level in the existing network locations are on average 0.7 km laterally and 2 km vertically. Correlation characteristics and relative location improvement are remarkably similar across most of Northern California, implying the general applicability of these techniques to image high-resolution seismicity caused by a variety of plate tectonic and anthropogenic processes. We show that consistent long-term seismic monitoring and data archiving practices are key to increase resolution in existing hypocenter catalogs and to estimate the precise location of future events on a routine basis.