Purpose: In order to build a realistic manual wheelchair simulator, an accurate dynamic model of the wheelchair-user system is needed. Besides, the characterization of this model necessitates the continuous knowledge of each caster wheel's orientation (CWO). The aim of this work is to propose a method that predicts these orientations based on the rear wheels' kinematics. Methods: A standard wheelchair equipped with two instrumented wheels was moved following straight and curvilinear patterns. A first-order model expressing the variation rate of the CWOs with respect to the rear wheels' kinematics was developed. The numerical integration of the model leads to an estimate of the CWOs. Simultaneously, the orientation of a caster wheel was measured using a three-dimensional optoelectronic system. Results: The estimated and the observed orientation values were compared. The estimation error was very low (<5 deg) when the orientation does not abruptly change, and stayed below 10 degrees at all times. Conclusion: Based on our data, it is possible to predict the CWOs with a fairly good accuracy using only the geometrical parameters of the wheelchair and the data from the instrumented rear wheels. This method is promising since it will help characterizing the wheelchair-user dynamic model in the future.