Reports by Nadeau and Dolenc (2005) that tremor had been detected near Cholame Valley spawned an effort to use UPSAR (U. S. Geological Survey Parkfield Seismic Array) to study characteristics of tremor. UPSAR was modified to record three channels of velocity at 40–50 sps continuously in January 2005 and ran for about 1 month, during which time we recorded numerous episodes of tremor. One tremor, on 21 January at 0728, was recorded with particularly high signal levels as well as another episode 3 days later. Both events were very emergent, had a frequency content between 2 and 8 Hz, and had numerous high-amplitude, short-duration arrivals within the tremor signal. Here using the first episode as an example, we discuss an analysis procedure, which yields azimuth and apparent velocity of the tremor at UPSAR. We then provide locations for both tremor episodes. The emphasis here is how the tremor episode evolves. Twelve stations were operating at the time of recording. Slowness of arrivals was determined using cross correlation of pairs of stations; the same method used in analyzing the main shock data from 28 September 2004. A feature of this analysis is that 20 s of the time series were used at a time to calculate correlation; the longer windows resulted in more consistent estimates of slowness, but lower peak correlations. These values of correlation (peaks of about 0.25), however, are similar to that obtained for the S wave of a microearthquake. Observed peaks in slowness were traced back to source locations assumed to lie on the San Andreas fault. Our inferred locations for the two tremor events cluster near the locations of previously observed tremor, south of the Cholame Valley. Tremor source depths are in the 14–24 km range, which is below the seismogenic brittle zone, but above the Moho. Estimates of error do not preclude locations below the Moho, however. The tremor signal is very emergent but contains packets that are several times larger than the background tremor signal and lasts about 5 s. These impulsive wavelets are similar to low-frequency earthquakes signals seen in Japan but appear to be broader band rather than just higher in low-frequency energy. They may be more appropriately called high-energy tremor (HET). HET signals at UPSAR correlate well with the record of this event from station GHIB of the HRSN borehole array at Parkfield and HETs typically have a higher cross-correlation coefficient than the rest of the tremor event. The amplitudes of a large HET are consistent with a magnitude of 0.1 when compared with a M2.3 event that had about the same epicenter. Polarizations of the tremor episode at UPSAR are mostly just north of east. Both linearity and azimuth evolve over time suggesting a change in tremor source location over time and linearity is typically higher at the HETs.