According to the massive and high-dimensional characters of the thermal parameter dynamic data derived from thermal power plants, a new fault diagnostic method for the thermal parameter sensor was proposed that was based on dynamic data mining. First, thermal parameter signals were decomposed into a series of intrinsic mode functions(IMFs) and a residual, through which the dynamic mining of sensor fault feature was effectively realized. Second, the variances of IMFs and the residual were proposed as eigenvectors for creating a Euclidean distance criterion function, and then a radical basis function(RBF) neural network was used to verify whether the sensor was at fault or not. Finally, based on expert experiences, the difference value between the measured and calculated values of the sensor were analyzed to classify the fault patterns. A simulative experiment was carried out based on the actual operation data of a 600MW thermal power plant unit. The calculation results verified that the proposed method can distinguish a signal change caused by a sensor fault from normal process dynamics only with the samples of the thermal parameter sensor in a normal situation, and can identify the working condition and fault patterns of the thermal parameter sensor quickly and accurately.