To reduce impacts of measurement errors on plant variables, data reconciliation is widely applied in process industries. Reconciled measurements are used in applications such as performance monitoring, process control, or real-time optimization. However, precise estimation generally relies on accurate and detailed process models which could be difficult to build in practice. The trade-off between estimate precision and model complexity is a relevant challenge motivating the development of effective observers with limited modeling efforts. This paper proposes a data reconciliation method based on a simple mass and/or energy conservation sub-model that also considers the autocovariance function of plant node imbalances. The observer is applied to simulated benchmark plants and its performance is evaluated in terms of variance reduction and robustness against modeling errors. Results show a superior performance in comparison with classical sub-model based methods and reveal less performance degradation than the Kalman filter in presence of model uncertainties. (C) 2012 Elsevier Ltd. All rights reserved.