It is necessary that a reliable and optimal control system has the ability of fault tolerance, which recovers the faulty and/or missed data in time. Principal component analysis (PCA) is presented to model HVAC monitored systems by using the measured data under normal operation condition (NOC). PCA splits the measurement space into two subspaces, one principal component subspace (PCS) and the other residual subspace (RS). When the faulty or missed data is observed, it will be projected into PCS and RS. It is then recovered by sliding the faulty or missed data to PCS via iteration. Examples of HVAC monitoring system have demonstrated that the approach has good performance to recover faulty or missed data and thus can be embedded into the system to achieve fault-tolerant control.