Cross correlation is applied to a variety of useful scientific applications. We have created a relocated double-difference earthquake catalog for northern California comprising 456,041 events and 48 billion correlation measurements. The average improvement in location errors is one to two orders of magnitudes (three orders of magnitude for repeats). For correlation detection, we achieve an order of magnitude improvement (one magnitude unit reduction in detection threshold). It is possible to detect aftershocks buried in the coda of a mainshock. Also a difference of 3.3 magnitude units between the master template and slave event can produce statistically significant detections. It has been discovered there is an abundance of repeating events in China (13%) and northern California (12%). They are clustered in time as well as space and able to assess location errors of routine catalogs (~15 km). About half are foreshocks. We have analyzed foreshock sequences for 612 M ≥ 4 mainshocks in northern California (ten times the number of mainshocks as in previous studies). A negative correlation of foreshock occurrence with depth is observed. There is a positive correlation of foreshock magnitude with mainshock magnitude (first time observed). One prediction of a pre-slip model is that foreshock magnitude may scale with nucleation region. We investigate the presence of preseismic velocity changes before large earthquakes. For an event in Korea using the doublet method on a rare foreshock sequence produces biases up to 1% velocity change. Fitting with a plane wave to correct delays due to slight position differences improves the measurements by an order of magnitude. The upper bound on preseismic changes ranges from 0.02% to 0.08%. Applying an ambient noise technique at Parkfield, CA, using all possible pairs (702) measures an upper bound on preseismic changes to be 0.026% for one day stacks or 17% of the -0.15% coseismic signal.