SUMMARYData from three-dimensional experiments performed on sand in true triaxial equipment have been reviewed to sort out apparent disarray resulting from their interpretation. This has been done based on analyses made possible by recent developments and understanding of factors influencing sand behaviour: occurrence of shear banding, boundary conditions and/or specimen slenderness ratio, cross-anisotropy, and stability of experimental technique. These factors are reviewed and test data from the literature are evaluated. Experimental data are divided into three groups in which: (a) homogeneous behaviour controls the sand strength; (b) shear banding affects the shape of the three-dimensional failure surface in the midrange of values of b ¼ ðs 2 À s 3 Þ=ðs 1 À s 3 Þ; and (c) the data has been misinterpreted. Appropriate interpretation of three-dimensional strength data for sand exhibiting isotropic, homogeneous behaviour is represented by a smoothly rounded triangular failure surface expressible in terms of the first and third stress invariants. Shear banding effects will cause the failure surface to be 'indented' in the midrange of b-values in all sectors of the octahedral plane. Effects of cross-anisotropy will result in lower strengths in sector III than in sector I of the octahedral plane, and the failure surface will appear as rotated around the stress origin in principal stress space.