We developed a new technique CAPloc to retrieve full source parameters of small seismic events from regional seismograms, which include origin time, epicenter location, depth, focal mechanism, and moment magnitude. Despite rather complicated propagation effects at short periods, a simple localized one-dimensional model can well explain signals of periods 3--10 s if we break the three-component records into different segments and allow differential time shifts among them. These differential time shifts, once established from a calibration process or a well-determined tomographic map, can be used together with P wave travel times to refine an event's location. In this study, we tested whether our new method could produce satisfactory results with as few as two stations, so that we can improve source estimates of poorly monitored events with sparse waveform data. We conducted the test on 28 events in the Tibetan Plateau. The focal mechanisms and locations determined from only two stations agree well with those determined from a whole PASSCAL array. In particular, our new method produces better locations than International Seismological Centre, with the average mislocation error reduced from ~16 to ~5 km. We also tested whether an event's depth and mechanism can be determined separately from its epicenter relocation in a two-step approach. We find that the two-step approach does not always give the correct solution, but the reliability of a solution can be evaluated using a reduced chi-square value.