A number of algorithms have been proposed to model the adaptive behavior of bone under load. However, the predictions of several models have neither been compared nor have they been systematically related to in vivo data. To this end, the stress states of loaded implant-bone interfaces were analyzed before and after osseointegration using finite element (FE) techniques. In a preliminary step, an FE mesh of a cylindrical implant encased in a cancellous core surrounded by a cortical layer was constructed, and the stresses and strains that developed at the interface were determined. The implant was loaded with 100 N vertical and 30 N lateral loads. Using this structure, the peak compressive and tensile stresses were determined. Then bone remodeling predictions were assessed using three different models: von Mises equivalent strain, strain energy density and effective stress. Finally, a systematic search of the literature was conducted to relate the numerical predictions to existing in vivo data. The FE simulations led to the following conclusions: (1) calculated compressive stresses were lower than the ultimate compressive stresses of cortical and cancellous bone. (2) Calculated tensile stresses were generally superior to experimental data on the tensile strength of the bone-implant interface. (3) With one exception, the predictions of all models were homogeneously grouped on the stimulus scales. (4) The predictions of the models as to bone gain or loss were not consistent and at times contradictory. It is hypothesized that this effect is linked to a lazy zone that is too narrow. With respect to the application of the numerical models to in vivo data, peak strains and strain energy densities were consistent with in vivo data. No in vivo data were found that supported effective stress as a stimulus.