Abstract: In structural health monitoring (SHM), the environmental effects such as the ambient noise and the variation in temperature will impede the accuracy of damage identification. So it is important to remove theses effects to decrease the uncertainty in damage detection results. In this paper, a new method based on wavelet packet energy transform and Probabilistic Principal Component Analysis (PPCA) is proposed to detect the damage based on the data under different temperatures and noise levels. The vibration data of the structure are decomposed into the wavelet packet components, and then the wavelet packet energy is calculated, which is an indicator of the structural damage for the wavelet packet energy is a damage-sensitive parameter. The PPCA is pursued to the wavelet packet energy index for dimensionality reduction and environmental effects elimination. The numeral example of Binzhou Yellow River Bridge is employed to illustrate the applicability of the method proposed in this paper.