Building utilization can be enhanced by tracking occupants. Non-intrusive tracking of occupants using floor-vibration measurements is beneficial for many smart-building applications such as energy consumption, security enhancement, and care-giving. Also, unlike cameras, vibration measurements do not undermine privacy. Current methodologies for interpreting floor vibrations are often data-driven and thus, do not account for the varying floor stiffness due to structural elements such as beams, columns and walls. In this paper, characteristics of floor vibrations are evaluated in order to assess their potential to perform model-based occupant localization. Empirical analyses of a full-scale case study are conducted to determine signal characteristics that provide information regarding occupant location. Structural elements are found to have a significant influence on measured floor vibrations due to footstep impacts. Moreover, the measured response is not a monotonic function of distance between the sensors and impact locations. Thus, the use of a model is necessary to interpret occupant locations. Using a sensitivity analysis, several signal characteristics are suggested for occupant localization. Using a physics-based model for interpreting measurement data is an ill-posed inverse task. Solutions to such tasks are sensitive to presence of uncertainties from multiple sources. Two sources of uncertainty measurement uncertainty and uncertainty from occupant gait, have been investigated. Incorporating these uncertainties improves occupant localization.