The measured fatigue strength of a material can be affected by specimen size:tests using a large stressed volume may show a low fatigue strength due to the increased probability of finding weak regions. A Weibull analysis revealed an important size effect in bone and predicted this effect with an accuracy of 12%. This approach also explained apparent inconsistencies in the published data and made it possible to separate and quantify the effects of frequency, loading mode, and material source. The effect of frequency is the same for human and bovine bone, and the differences between different types of loading (tension, compression, and bending) are small (maximum: 12%). By extrapolating to the volume of whole bones, it is concluded that large bones will have a fatigue strength much lower, by a factor of 2-3, than that measured by conventional tests. Failure within 10(5) cycles is expected to occur at cyclic stresses of 23-30 MPa in human long bones and of 32-43 MPa in bovine bones. Repair is therefore needed to prevent failure at physiological stress levels.