A fault trend prediction problem for a class of exponential fault process is studied. Under the condition that the measure variable is disturbed by a stationary noise, the fault process is detected by statistical analysis of the measurement firstly. According to the model assumption, the parameters of the fault trend process can be obtained by using STF ( strong tracing filter). After the extraction of trend component, the modeling error series becomes a stationary series, which can be used for normal ARMA time series analysis. Finally, the whole prediction can be acquired by combining trend prediction and time series analysis. Computer simulations validate the effectiveness of the proposed method.