AbstractWe employed our new polyaxial cell to carry out true triaxial compression tests on dry (jacketed) rectangular prisms of two crystalline rocks, in which different magnitudes of the least and intermediate principal stresses σ3 and σ2 were maintained constant, and the maximum stress σ1 was increased to its peak level in strain control. Both Westerly granite (Rhode Island, USA) and KTB amphibolite (Bohemian Massif, Germany) revealed similar mechanical behaviour, much of which is missed in conventional triaxial tests in which σ2 = σ3. Compressive failure in both took the form of a main shear fracture, or fault, steeply dipping in the σ3 direction. Compressive strength rose significantly with the magnitude of σ2, suggesting that the commonly used Mohr-type strength criteria, which ignore the σ2 effect, predict only the lower limit of rock strength. The true triaxial strength criterion for each of the crystalline rocks can be expressed as the octahedral shear stress at failure as a function of the mean normal stress acting on the fault plane. We found that the onset of dilatancy increases considerably for higher σ2. Thus, σ2 extends the elastic range for a given σ3 and, hence, retards the onset of the failure process. The main fracture dip angle was found to increase as σ2 rises, providing additional confirmation of the strengthening effect of σ2. SEM inspection of the micromechanics leading to specimen failure showed a multitude of stress-induced microcracks localized on both sides of the through-going fault. Here too the effect of σ2 is noted, in that microcracks gradually align themselves with the σ1-σ2 plane as the magnitude of σ2 is raised.