Griffith's theory of brittle fracture postulates that natural solid materials contain flaws that provide local sources of tensile-stress concentration, even when applied load is compressive. Fracture is predicted to ensue when the maximum tensile-stress concentration equals bond strength. This theory has proven to be fundamental in tensile fracture of isotropic brittle materials, and it appears to be fundamental in brittle shear fracture of rocks. Griffith assumed that the longitudinal cross-section of an elongate flaw could be represented mathematically by an excentric ellipse for analysis by the theory of elasticity. Photoelastic experiments support this assumption, provided the ellipse has the same length as the flaw and the same radius of curvature at points of maximum tension as those at the flaw boundary. The theory is extended mathematically to predict initial direction of propagation of such flaws or Griffith cracks. Unlike pure tensile fracture, an isolated critical Griffith crack ceases crack growth rapidly under applied compression by attaining stable configuration. Hence, interaction between several cracks apparently is required to form the shear-fracture surface. Photoelastic experiments indicate that coalescence of neighbouring cracks may not occur for crack separation greater than a few crack lengths, unless applied stress is appreciably higher than that needed to initiate growth, and that certain en echelon arrays of Griffith cracks are more favourable for coalescence than others. Brace (1964) reports that straight segments of grain boundaries, in crystalline rocks tested, behave like Griffith cracks; those in en echelon array are apparently activated preferentially to form the shear-fracture surface.