This paper presents a practical approach to combine model-based fault detection with an adaptive threshold. The suitability of the proposed technique is illustrated through its application to the condition monitoring of a nonlinear electro-hydraulic plant. The paper begins by outlining the difficulties associated with modelling the plant and the steps taken to identify the uncertain factors that influence the accuracy of the resulting model. A linearised model is applied in this study. The reason for this is because of the availability of many well-developed model-based approaches and model parameter estimation techniques for linear systems. The errors due to the linearisation and stochastic factors are studied both experimentally and theoretically and are compensated for by using an adaptive threshold. The combined linearised model-based approach and adaptive threshold is not only easy for on-line implementation but also takes into account the unknown influences such as model errors, measurement noise, temperature fluctuation and hence leads to a reliable fault detection scheme. The performance of the proposed fault detection scheme is demonstrated in detecting several different fault types associated with the control components, actuator and sensor.