Accuracy of online measurement data is usually not satisfactory for coal-fired power plants due to constraints of measurement techniques. Data reconciliation can help to improve the accuracy of online measurement with no extra requirements on equipment upgrading. Traditionally, data reconciliation is applied to a sub-system of a coal-fired power plant, for instance, mass balance of a steam turbine system. In this work, we present a systematic approach where data reconciliation is applied to the overall thermal system of a real-life steam turbine power plant. Improvement of data accuracy is obtained via the proposed approach compared with sub-system studies. Optimization of system selection and configuration of a data reconciliation problem is analyzed. Results show that uncertainty of on-line measured data can be reduced by up to 50% in a 1000 MW ultra-supercritical coal-fired steam turbine power plant compared with previous studies. (C) 2016 Elsevier Ltd. All rights reserved.