We have conducted a series of fatigue tests on samples of bovine compact bone loaded in cyclic torsion. The fatigue strength (i.e. the range of stress needed to cause failure in a given number of cycles) was found to be lower than the fatigue strength of the same material in compression by more than a factor of two. We also tested intact chicken metatarsals and found a similar reduction in strength compared to compression testing of chicken tibiae. These results were predicted using a theoretical model in which fatigue failure was assumed to be dependant on the growth of microcracks, oriented approximately parallel to the bone's longitudinal axis but having misorientation angles of up to 30°. An effective stress range was derived which is a function of the normal and shear stresses, and thus of the Mode I and Mode II stress intensities experienced by the crack. These results may have important consequences for the understanding of fatigue in bone in vivo; relatively small amounts of longitudinal shear stress, which are often ignored in analysis, may contribute significantly to fatigue failures. This may shed light on the phenomenon of stress fractures and on the need for repair and adaptation in living bone.