The fidelity of MPC models can be significantly compromised if the effect of unmeasured disturbances is contained in the data set used to identify the process models: This paper addresses an asymptotical detection method (ADM) based on the normalized primary residual and the chi(2) statistic. An iteratively procedure, based on ADM, is proposed for removing data containing the effect of unmeasured disturbances until all the data passes the ADM chi(2) test. A detailed simulation of a refining FCC unit is used to evaluate the performance of the proposed method for identifying unmeasured disturbances. A series of random walk unmeasured disturbances with varying amplitudes was applied during a simulated testing period. The proposed algorithm was applied to iteratively identify and remove the corrupted data. After the corrupted data were removed, the identified MPC models showed significant improvement. The sensitivity of the ADM to measurement noise is also evaluated.