The modified Griffith fracture initiation criterion and other failure criteria were used, from the late 1960 to the present for predicting, or attempting to, the extent and orientation of fractures in the hanging wall of deep level stopes. Because of the apparent problematic nature of solving the fracture propagation part of the criteria, the search for alternative failure criteria were then pursued along the empirical route of which the now in common use Hoek-Brown criterion was developed. The extent of fracturing is predicted but the lack of predicting the orientations and extension of induced fractures around underground excavations prompted the re-examination of the failure process. It was found that a failure criterion requires the input of the strength, the orientation of the maximum principal stress as well as the associated orientation of the newly created macro-crack. The paper describes a theoretical framework by which it is possible to determine the data required for fracture coalescence as well as proposes an associated empirical failure criterion. The results obtained using the proposed failure criterion compare well with results obtained using the Hoek-Brown criterion while the fracture propagation model predicts the orientation of observed fracture patterns. The use of the methodology is illustrated by comparing FLAC generated fracture zones for a circular specimen, a pyroxenite pillar and a preliminary comparison with measured stope fractures.