We present our framework for monitoring and diagnosis. The framework is based on interaction between a qualitative diagnosis engine and a monitoring component that performs a signal to symbol transformation on the signals. We have applied the methodology to a real system, the cooling system of an automobile engine on which we have installed thermocouples and pressure sensors. Faults can be introduced into the cooling system in a controlled manner. We show that a combination of linear approximation techniques and statistical signal processing can provide robust symbolic signal values for the diagnosis algorithms. 6 Pages