I relocated two microearthquake clusters induced by hydraulic stimulation of a hot-dry-rock geothermal reservoir in the Rhine Graben near Soultz-sous-Forêts, France. The two clusters were chosen from a collection of clusters identified within a 16,000-event data set obtained in 1993. I determined P- and S-wave arrival times, manually, choosing distinctive peaks if waveforms were similar at a given station, or first breaks if waveforms were emergent or nodal, judging weights accordingly. Requiring a minimum of five arrival times yielded populous clusters of 226 and 355 events. In both cases, relocation revealed two distinct planar patches of activity, 100 to 200 m across, intersecting and truncating one another along a common edge. Linear segments traverse the planar patches and may mark intersections with seismically inactive fractures. The earliest activity occurred along these segments. These segments may therefore mark permeable zones that could be important components of the fluid-flow network. The interpretation of linear segments as flow paths is supported by path dimension constraints derived from fluid-flow modeling studies and by the expected confining effects of hydrothermal sealing found in cored fractures. Orientations of the planar features are verified through consistency with orientations found in core and logging studies. In addition, focal mechanisms obtained by constraining slip planes parallel to active planes indicate normal to right-lateral strike slip, consistent with the measured stress field. Focal mechanisms indicate mass deficits at inside corners, implying that the concentrations of activity at fracture intersections may result from slip on one fracture causing extension and a resulting permeability increase in the other. The improved location precision results from a ten-fold improvement in the precision of arrival-time estimation for phases that were similar, as estimated from the standard deviation of residuals, grouped by station and phase, before and after repicking. Only a two-fold improvement was noted for poorly correlated or nodal phases. For this data set, the manual phase picking yielded more interpretable detail in the location patterns, compared to results of an automatic, cross-correlation technique. The manual results could be used to guide improvements to automatic techniques to obtain precise locations of the entire data set.