Sensor failures are a major cause of concern in engine-performance monitoring as they can result in false alarms and, in some cases, lead to the condemnation of a non-offending component or section of the engine. This condition has the potential to increase engine downtime and thus incur higher operational costs. The fact that more than a single sensor could be faulty simultaneously should also not be overlooked. In this paper, we present a set of neural networks, modularly designed to diagnose and quantify single and dual-sensor faults in a two-shaft stationary gas-turbine. A further outcome of the analysis is the restructuring of the faulty data to a fault-free form through the filtering out of noise and bias. This restructured data can be used to perform sensor-based calculations accurately. The engine chosen for this analysis is thermodynamically similar in performance to the Rolls Royce (RR) Avon. The data used to train the networks were derived from a non-linear aero-thermodynamic model of the engine's behaviour. The results obtained show the good prospects for the use of this technique.