A novel method which uses wavelet transforms to detect and diagnose the abrupt fault of sensor by analysing the output signal was described. The paper also deduced the wavelet transforms of the normal signal outputs in sensor and five kinds of model abrupt fault signals. The abrupt faults of sensor were localized by detecting the singularity of the wavelet transforms signal. The characteristic vector matrix connecting with the types of faults was obtained by the energy distribution differences of the model fault signals on all decomposed wavelet scales. The sensor faults were classified according to the characteristic vector matrix. Finally, the simulation results of silicon pressure sensors verified the validity of the proposed method.