A model-based estimation technique is proposed to estimate the wheel-rail lateral forces and yaw moments of heavy haul locomotives for condition monitoring, based on discretetime Kalman filter theory. Four Kalman filters, involved individually with the single wheelset (SW) model, bogie model, half-vehicle (HV) model and full-vehicle (FV) model, are developed and compared. For validation of the present estimators, a multi-body system (MBS) dynamics model, accounting for significant DOFs and non-linearities, is established in SIMPACK. Field ring-line experiments of an HXN5 locomotive, with instrumented wheelsets for measuring the wheel-rail vertical and lateral forces, are carried out to validate the MBS model. Numerical examples are given by the present algorithms for estimation of the lateral forces and yaw moments of the HXN5 locomotive in high-, mild-and low-friction contact conditions, corresponding to dry, wet and leaf-covered tracks. It is demonstrated by comparison that the estimated results of the bogie, HV and FV filters are in better agreement with the predictions of the MBS model than those of the SW filter, and the dual-filter using a bogie or HV filter is more efficient than the FV filter.