Two mining induced tremors that occurred in a South African deep‐level gold mine are considered, with the aim to simulate the shear slip area and source mechanism using static boundary element numerical modelling. Large seismic tremors, induced by the extraction of the tabular ore body typically 1 m high over horizontal spans of kilometres, are a major seismic hazard in South African gold mines. For the cases considered here, virgin stress levels are ∼54 MPa, and seismic related damage is mainly due to geological faults failing in shear under gravitational loading. The tremors occurred ∼2 km below surface, 700 m apart in space, and in consecutive months. Seismic recordings and underground observations suggested that the tremors, of local magnitudes 3·0 and 4·0 respectively, were associated with normal slip on similar geological faults. A boundary element numerical model was implemented to some degree of accuracy, since the fault geometries could be inferred from extensive mining spans and fault intersections, and an in situ stress state could be estimated from other available information. Furthermore, the rock mass stress changes and associated surface stress changes on faults could be modelled from monthly measured advance of the mining excavations. Simulation of the seismic sources entailed finding appropriate Mohr–Coulomb strength parameters to allow yielding of the faults at the actual mining stage, and quantitatively in agreement with the seismic moment as inferred from seismic waveform recordings. A critical input into the model was the use of non‐zero cohesion in addition to friction angle as specified in the Mohr–Coulomb shear failure criterion, allowing the simulation of coseismic slip. Hence, the breaking of fault cohesion and subsequent stress transfer to the surrounding fault area is simulated, albeit statically, conforming to the perceived triggering of mining related seismic tremors by induced stress changes. The back analysis furthermore suggested that coseismic stope closure can comprise a significant part of the total recorded seismic moment, depending on the proximity of the hypocentre to mining excavations. From the back analysis a modelling methodology and input parameters are proposed, aiming at identifying vulnerable mining excavations subject to seismic hazard posed by geological structures, hence providing relevant knowledge towards safer mining.