The QuakeFinder network of magnetometers has recorded geomagnetic field activity in California since 2000. Established as an effort to follow up observations of ULF activity reported from before and after the M Combining double low line 7.1 Loma Prieta earthquake in 1989 by Stanford University, the QuakeFinder network has over 50 sites, fifteen of which are high-resolution QF1005 and QF1007 systems. Pairs of high-resolution sites have also been installed in Peru and Taiwan. Increases in pulse activity preceding nearby seismic events are followed by decreases in activity afterwards in the three cases that are discussed here. In addition, longer term data is shown, revealing a rich signal structure not previously known in QuakeFinder data, or by many other authors who have reported on pre-seismic ULF phenomena. These pulses occur as separate ensembles, with demonstrable repeatability and uniqueness across a number of properties such as waveform, angle of arrival, amplitude, and duration. Yet they appear to arrive with exponentially distributed inter-arrival times, which indicates a Poisson process rather than a periodic, i.e., stationary process. These pulses were observed using three-axis induction coil magnetometers that are buried 1-2 m under the surface of the Earth. Our sites use a Nyquist frequency of 16 Hertz (25 Hertz for the new QF1007 units), and they record these pulses at amplitudes from 0.1 to 20 nano-Tesla with durations of 0.1 to 12 s. They are predominantly unipolar pulses, which may imply charge migration, and they are stronger in the two horizontal (north-south and east-west) channels than they are in the vertical channels. Pulses have been seen to occur in bursts lasting many hours. The pulses have large amplitudes and study of the three-axis data shows that the amplitude ratios of the pulses taken from pairs of orthogonal coils is stable across the bursts, suggesting a similar source. This paper presents three instances of increases in pulse activity in the 30 days prior to an earthquake, which are each followed by steep declines after the event. The pulses are shown, methods of detecting the pulses and calculating their azimuths is developed and discussed, and then the paper is closed with a brief look at future work.