Failure data from compression and extension testing of Dunnville sandstone and Webatuck dolomite were analyzed using three different criteria: Mohr-Coulomb, Hoek-Brown, and Paul-Mohr-Coulomb. An emphasis was placed on the details of the failure criteria under tensile-compressive loading conditions, which was possible due to a dog-bone specimen geometry. Tension cut-offs were implemented when using Mohr-Coulomb and Paul-Mohr-Coulomb due to unrealistic predictions of (theoretical) tensile strength. It was determined that under tensile-compressive loading conditions, a tension cut-off provided a better match than Hoek-Brown to the data for the two rock types examined. Hoek-Brown overestimates the uniaxial tensile strength by a factor between 2-3 and has the greatest overall least absolute residual error in the low mean stress region.