The quality of on-line measured operational data is usually not satisfactory for the performance monitoring of coal-fired power plants, due to the low accuracy of measuring instrument. Data reconciliation is a data preprocessing technique which can improve the accuracy of measured data through process modeling and optimization, and can also be used for gross error detection together with a statistical test method. In this work, we provide a mathematical framework for gross error detection in power plants via data reconciliation. We also provide case studies to implement the proposed framework in the feed water regenerative heating system of a real-life 1000 MW ultra-supercritical coal-fired power generation unit. Data reconciliation simulation results show that the relative root mean squared errors of the primary flow measurements, namely the outlet flow rate of the #1 feed water heater, the outlet flow rate of the feed water pump, and the inlet flow rate of condensate water in the deaerator are reduced by 72%, 40%, 22%. Simulation results also show that data reconciliation is effective for accuracy improvement when estimated error standard deviations are different from the actual ones and when random errors follow generalized normal distributions. We then provide a case study where gross error detection is performed together with a global test and a serial elimination strategy, and a gross error in the measurement of outlet flow rate of the feed water pump is successfully detected and validated by the on-site inspection and maintenance records of the power plant. (C) 2014 Elsevier Ltd. All rights reserved.