Primary stability is recognised as an important determinant in the aseptic loosening failure process of cementless implants. An accurate evaluation of the bone-implant relative micromotion is becoming important both in pre-clinical and clinical studies. If the biological threshold for micro-movements is in the range 100-200 micrometer then, in order to be discriminative, any method used to evaluate the primary stability should have an accuracy of 10-20 micrometer or better. Additionally, such method should also be able to report the relative micromotion at each point of the interface. None of the available experimental methods satisfies both requirements. Aim of the present study is to verify if any of the current finite element modelling techniques is sufficiently accurate in predicting the primary stability of a cementless prosthesis to be used to decide whether the micromotion may or may not jeopardise the implant osseointegration. The primary stability of an anatomic cementless stem, as measured in vitro, was used as a benchmark problem to comparatively evaluate different contact modelling techniques. Frictionless contact, frictional contact and press-fitted frictional contact conditions were modelled using alternatively node-to-node, node-to-face and face-to-face contact elements. The model based on face-to-face contact elements accounting for frictional contact and initial press-fit was able to predict the micromotion measured experimentally with an average (RMS) error of 10 micrometer and a peak error of 14 micrometer. All the other models presented errors higher than 20 micrometer assumed in the present study as an accuracy threshold.