SUMMARY We test three different methods for selecting the fault plane from the two nodal planes in an individual focal mechanism, which are implemented in a linear stress inversion technique, to constrain the geometry of subsurface seismogenic faults. The three fault-selection methods use the misfit angle between the slip orientation and maximum shear stress direction on the nodal planes, fault instability and slip tendency, respectively, to select the fault planes. These three fault-selection methods are applied to various types (synthetic, simulated and real) of focal mechanism data sets where the fault planes are already known. In particular, synthetic focal mechanism data sets are generated with the assumption of different levels of pore pressure variation (up to 60 per cent of the minimum principal stress) that may be involved in activating faults with diverse orientations. The instability method performs the best among the three fault-selection methods, with 60–100 per cent of the faults correctly selected in each of the tested data sets. The slip-tendency method is slightly less accurate than the instability method, especially when the earthquakes occur in a highly variable pore pressure environment, and the misfit-angle method is relatively ineffective in selecting the faults, especially in noisy data sets. The test results show that the instability method is the most effective in correctly selecting the faults when the instability of the selected fault plane is significantly higher than that of the auxiliary plane, whereby the instability ratio of the selected (fault) plane to the auxiliary plane is above ∼1.4; this constraint can improve our ability to identify subsurface seismic faults. We apply this stress inversion technique, which implements the instability method, to the 2016–2017 induced earthquakes in Pohang, South Korea. We invert 53 well-constrained, well-located focal mechanism solutions to derive the stress condition, with the nodal planes possessing a higher instability selected as faults. The earthquakes, which occurred in spatially distinct areas of the region, have been associated with water injection into two boreholes (PX-1 and PX-2). Approximately 70 per cent of the identified faults for the PX-2-related earthquakes are well aligned in terms of both their locations and orientations, thereby indicating that these earthquakes occurred along a single, large-scale fault. The fault planes with instability ratios above ∼1.3 are all correctly selected for the PX-2 fault. Although there are more variations in the identified fault orientations of the PX-1-related earthquakes, some of the fault planes with high instability ratios (>1.4) are generally subparallel to one another. Both the locations and orientations of these high-instability-ratio planes are well aligned, suggesting the presence of a large-scale fault that is subparallel to the PX-2 fault. This study demonstrates the potential of effectively identifying and imaging subsurface seismic faults using only information on fault mechanics (i.e. stress and focal mechanisms).