When measuring the internally deposited activity in the bone of a subject, the placement of the detector is critical. This study reports the simulated counting efficiencies for three counting geometries, the skull, knee and shin, using 13 different voxel phantoms. It shows that the range of counting efficiencies for a given geometry is large for the studied phantoms, especially at low energies. Skull counting offers higher efficiency for low energies such as the 17 keV compared to knee counting or shin counting, but this advantage disappears when the energy is higher such as at 185 keV. This work also shows that the calibration phantom may greatly impact the accuracy of the activity estimate in bone counting, with uncertainties increasing greatly as the photon energy is reduced. Estimating the activity of a radionuclide in bone from direct counting has large uncertainties, and the dose calculated from a skeleton measurement would need careful analysis and, if possible, supporting data from other bioassay measurements.