In this report we discuss the setting of the San Andreas Fault Observatory at Depth (SAFOD) and the interpretation of downhole microearthquake recordings of fault zone guided (Fg) waves and P-wave tomography. These observations reveal a new, seismologically significant branch in the San Andreas Fault (SAF) zone at the SAFOD site.
 SAFOD is located approximately 14.5 km northwest of Parkfield, California, and 1.7 km southwest of the surface trace of the SAF (Figure 1). At this location, the SAF separates Pacific Plate Salinian rocks from North American Plate Franciscan mélange [Dibblee, 1971]. In addition, it coincides with the transition from the creeping to the locked sections of the fault [e.g. Nadeau and McEvilly, 1997]. SAFOD's goal is to investigate the structure and mechanics of this unique fault section, and the small, repeating earthquakes that take place there (“target” earthquakes).
 The observatory currently consists of two boreholes: 1) A vertical, 2.1 km deep Pilot Hole (PH), drilled in the summer of 2002 and operated as a seismic monitoring station and instrument test bed [Chavarria et al., 2004], and 2) A deviated Main Hole (MH), initially drilled in the summer of 2004 to a depth of ∼2.5 km and continued in the summer of 2005 to a depth of ∼3.1 km. Following completion of the PH, a 32-level array of 3-component, 15-Hz seismographs spaced 40 meters apart was installed in the deepest portion of the hole. The array recorded microearthquakes and surface explosions until July 2004; a smaller array, consisting of the original upper seven levels, collected data from July 2004 until April 2005.
 SAFOD investigators have studied the PH data intensively in an effort to locate the target earthquakes [Roecker et al., 2004; Thurber et al., 2004; Waldhauser, 2001]. We used the PH data to produce a P-wave velocity cross section through the drill site and SAF (Figure 2). The cross-section was based on data from a 2003 refraction line along this same section [Chavarria et al., 2004; Roecker et al., 2004] (http://quake.wr.usgs.gov/research/parkfield/2003site.html). These refraction data, plus network calibration shots and nearby microearthquakes, have also been used to study seismic waves scattered from the local fault structure [Chavarria et al., 2003; Chavarria, 2004]. These data show that a zone of low seismic velocities exists between the drill pad and the SAF (Figure 2 and Hole et al., 2006).
 Starting from the PH pad, the 2004 MH was bored vertically for more than one kilometer. Below 1 km, the borehole was deviated ∼55° from vertical to steer toward the target earthquakes, which are thought to lie below the surface trace of the SAF. The 2004 drilling stopped ∼0.7 km southwest of the surface trace, where two features coincide: 1) A highly fractured lithological contact identified in a bottom-hole core [Solum et al., 2005] and 2) The southwestern edge of a structure which was imaged using scattered microearthquake waves recorded on the PH array (Figure 3). The MH seismometer was installed at the base of the hole.
 During the six weeks that the MH seismometer was operating, it recorded 363 events on the SAF that are included in the Northern California Earthquake Catalog (http://www.ncedc.org/ncedc/catalog-search.html). Of these, 126 contain clear examples of Fg-waves (Figures 4a and 4b) [Li et al., 2004; Fohrmann et al., 2004]. These signals are thought to propagate within the low-velocity fault cores. They are characterized by high amplitude, low frequency, dispersive waves, which follow the S-wave coda and result from constructive interference of critically reflected waves [Li et al., 1990; Ben-Zion and Aki, 1990]. Large Fg-waves were observed for sources lying beyond ∼2 km northwest and 3 km southeast of the MH seismograph (Figure 5a).
 The fault guided waves may be generated by the scattering of microearthquake body waves from events beyond these points (the “site-effect” model of, for example, Lewis et al., 2005). The alternative is the classic wave guide model of Fg-waves: Signals can only be observed when the source and receiver are in the same (more-or-less) uniform and continuous fault segment and have an approximately straight line source-to-receiver path. By this model, in order to produce the Fg-waves that were recorded by the MH seismometer, the newly discovered fault must form a continuous plane connected with the active SAF northwest and southeast of the seismometer.
 The new fault must be connected to the main SAF below the depth of events along the 5 km segment not producing Fg-waves. A fault connected above these earthquakes would block Fg-waves with a crooked source-to-receiver path between the events and seismemoter (Figure 5B). This is a likely possibility because the SAFOD site is situated on the Middle Mountain flower structure segment of the SAF, characterized by upwardly divergent fault splays as is typical of strike-slip fault zones [Woodcock and Fischer, 1986]. It is not surprising that this branch has not been identified until now, rapid uplift has left much of Middle Mountain's flanks covered with landslides, thus burying the fault's surface trace. Both the new fault discussed here and the (presumably) main fault below the surface trace of the SAF must cut through the Middle Mountain Syncline. These faults may trap a sliver of sediments related to the Franciscan mélange (eastern) side of the SAF zone. This inference is consistent with potential field models of the rocks between the SAFOD drill site and the SAF surface trace [McPhee et al., 2004]. Our P-wave velocity tomography suggests that these sediments most likely extend to at least ∼1.5 km, deeper than indicated in the work of McPhee et al. . In fact, the newly discovered fault could be an older trace of the SAF, separating the Franciscan mélange or other pre-tertiary sediments from the Pacific Salinian rocks. The main surface trace of the SAF may thus be one of many active branches. It is also possible that the new fault may represent the actual Pacific - North American plate boundary.
 Regardless of interpretation, the presence of a major wave-guiding fault must be taken into account when interpreting the geology of the SAFOD site and the mechanics of the local San Andreas Fault system. Further exploration and mapping of the fault structure will be possible in 2007 when multilateral cores are drilled.