This paper presents a complete methodology for detection and diagnosis of faults in variable air volume air handling units. Three cases are considered: (a) an off-line fault detection technique for existing buildings, (b) an automatic on-line fault detection technique for integration in building management systems (BMSs) of upcoming not very complex buildings and (c) an automatic on-line fault detection as well as diagnosis technique for BMSs of upcoming complex automated buildings. The method is based upon the auto regressive exogenous model and recursive parameter estimation algorithm. The proposed model and methodology have been trained by using several days of normal real time operational data and validated on data obtained by introducing faults artificially under normal operating conditions. It is concluded that the method is robust and can detect faults in dampers, sensors and PID control.