It is beneficial for the safe and optimal operation of system and energy conservation to find out and eliminate the faults existing in HVAC systems in time. For a large-scale and complex HVAC system, an automatic fault detection and diagnosis system is needed to ensure it to operate safely and reliably. Principal component analysis (PCA) approach has been used to detect and recover sensor faults in central chilling system through simulation data. However, it is found that sensor validation index (SVI) of PCA cannot identify the flow meter fault in the system. This paper explains the phenomenon in terms of the collinearity among flow meters in central chilling system. A novel approach, wavelet transform, is presented to identify the faults of flow meter as a substitute for SVI of PCA. Because of the good property of wavelet in local time-frequency, the approach can detect various variations in a sensor signal, such as ramp, step, and discontinuity. Thus, it is capable to identify the faults of flow meter in central chilling system. Some examples are given to show its ability of fault isolation for flow meters.