TRANSCEND, our system for fault detection and isolation of complex dynamic systems, uses a model based approach to predict and analyze transient effects resulting from abrupt faults in the system. Abrupt faults are attributed to discrete and persistent parameter value changes. Fault isolation is performed by matching features extracted from the transients against those predicted by the model. This paper discusses a statistical signal processing approach to transient detection and analysis using a time-frequency representation of the signal. The approach is robust for the detection task and it provides feature values for the initial fault isolation steps.