Twelve moderate‐magnitude earthquakes (mb 4–5.5) in the Pamir‐Hindu Kush region are investigated to determine their focal mechanisms and to relocate them using their regional waveform records at two broadband arrays, the Kyrgyzstan Regional Network (KNET), and the 1992 Pakistan Himalayas seismic experiment array (PAKH) in northern Pakistan. We use the “cut‐and‐paste” source estimation technique to invert the whole broadband waveforms for mechanisms and depths, assuming a one‐dimensional velocity model developed for the adjacent Tibetan plateau. For several large events the source mechanisms obtained agree with those available from the Harvard centroid moment tensor (CMT) solutions. An advantage of using regional broadband waveforms is that focal depths can be better constrained either from amplitude ratios of Pnl to surface waves for crustal events or from time separation between the direct P and the shear‐coupled P wave (sPn + sPmP) for mantle events. All the crustal events are relocated at shallower depths compared with their International Seismological Centre bulletin or Harvard CMT depths. After the focal depths are established, the events are then relocated horizontally using their first‐arrival times. Only minor offsets in epicentral location are found for all mantle events and the bigger crustal events, while rather large offsets (up to 30 km) occur for the smaller crustal events. We also tested the performance of waveform inversion using only two broadband stations, one from the KNET array in the north of the region and one from the PAKH array in the south. We found that this geometry is adequate for determining focal depths and mechanisms of moderate size earthquakes in the Pamir‐Hindu Kush region.