A study on the application of a direct virtual sensor-moving horizon estimator (DVS-MHE) to the problem of vehicle side-slip angle estimation is carried out. In particular, it is shown that a stable MHE can be represented as nonlinear finite impulse response (NFIR) filter. Then, in order to allow online implementation and guaranteed estimation accuracy, an optimal NFIR filter is derived directly from the data by means of a DVS approach. Comparisons between the standard model-based MHE approach and the DVS approach are carried out using a detailed vehicle model.