An improved MT-NT combined method of gross error detection and data reconciliation for chemical process data is presented. The proposed method overcomes the decrease of the coefficient matrix rank by using a strategy of detecting and reconciling via successive iteration. This method is efficient in detecting gross errors, reducing operation as well as avoiding loss of useful information. The case study shows the improved performance compared to the existing MT-NT method. (C) 2004 Elsevier Ltd. All rights reserved.