Model-based fault detection and isolation (FDI) are required to assess plant measurements before using them for data reconciliation, process observation, control and optimization. Frequently, measurement gross error detection methods are performed making use of steady-state mass or energy conservation equations, but neglecting process variability. To take account of the process dynamics, FDI techniques are available which requires a full phenomenological model of the plant. This paper proposes an intermediate FDI method that re-establishes time coherence between process variables, using linear empirical models of plant node dynamics and stationary conservation equations to generate node imbalances, which are residuals in the parity space. The performance of the method is assessed for different fault types (sensor biases or process leaks) and compared to the performance of the steady-state and stationary node imbalance methods, as well as the full dynamic approach. The method is illustrated for a mineral separation plant. The principal advantage of the proposed method is its ability to efficiently detect faults during process transitions, even when the dynamic model is approximately known. (C) 2006 Elsevier Ltd. All rights reserved.