In order to improve the fault diagnosis theory of switched systems, observer-based method is proposed to deal with the fault diagnosis issue subjected to the parameter uncertainty, unknown process disturbances and unexpected faults. By employing proportional-integral observer (PIO) technique and the average dwell-time (ADT) method, an H∞/H− performance level is considered to ensure the worst case of robustness to the disturbance and sensitivity to the fault simultaneously. Moreover, the obtained fault diagnosis observer can give accurate fault estimation online for the piecewise constant fault. Finally, a two-tank system example is introduced in the end to show the effectiveness of the proposed results.