This paper presents a new approach to the design of a robust observer-based fault detection scheme for diagnosing incipient faults, called the H∞/ μ robust fault detection observer (RFDO) which takes into account of the robustness against disturbances and sensitivity to faults simultaneously. The approach has originated from the robust H∞/ μ estimator which minimises the effect of disturbance on the estimation error and subsequently on the diagnostic residual. The effect of faults on the diagnostic residual is maximised by the proper selection of the performance bound and the estimation weighting matrix of the H∞ robust estimator. Depending on the class of uncertainty modelling considered, one or two Riccati equations are required for the H∞ estimator design. The H∞ estimator can be designed to be robust against; disturbance only, disturbance and parameter uncertainty only, and disturbance and a large variety of modelling errors. The μ robust estimator assumes a block diagonal structure for uncertainty and can produce less conservative design. The approach has been applied to a third-order system example chosen to demonstrate special design features. The results show that the fault detection scheme can detect incipient faults effectively even in the presence of disturbances and modelling errors.