The design of linear virtual sensors to estimate yaw rate for vehicle stability control systems is investigated. Standard model-based virtual sensor design techniques are compared to novel direct virtual sensor (DVS) design methodologies. The obtained DVS is stable and it can be used in a large range of operating conditions. It is shown how the use of virtual sensors derived directly from data and a suitable choice of the measured variables in sensor design improves the estimation and control accuracy. The effectiveness of the proposed DVS design is demonstrated by its employment in an existing yaw rate feedback loop, based on an active front steering actuator and designed using internal model control techniques. Robust stability is guaranteed in the presence of model uncertainty and of the DVS. In particular, the presented study shows that the DVS technology can be conveniently taken into account to replace physical sensors to obtain low cost stability control solutions for application on A and B segment cars.