A fault tolerant measurement system for a gas turbine in a combined cycle power plant, based on dynamic models, principal component analysis (PCA) and Q test, is presented. The proposed scheme makes use of a model-based symptom generator, which delivers fault signals obtained by using direct identification of parity relations and structured residuals. Symptoms are then analyzed in a statistical module achieving fault diagnosis and reconstruction of the faulty signals. The scheme presents as main advantage the ability of detecting faults in both input and output sensors due to its particular structure. Tests carried out on the gas turbine of the San Isidro combined cycle power plant in the V Region, Chile, show that Takagi-Sugeno fuzzy models present the best fitting performance and an acceptable computational cost in comparison with autoregressive exogenous. state space, and neural models. Real time software based on this scheme has been developed and connected to Osisoft PI System'. The software is running at Endesa Monitoring and Diagnosis Center in Santiago. Chile. (C) 2011 Elsevier B.V. All rights reserved.