Along with the increase of scale of hydropower station and single unit capacity, the vibration problems of hydropower houses are becoming acute. It is useful to assess a hydropower station safe, if the structural vibration characteristics of a hydropower house can be known. In this paper, basing on correlation analysis of the prototype observation vibration data of a huge underground power plant, the vibration prediction model of the hydropower house was established using relevance vector machine method (RVM). With the model, the vertical vibration responses of the powerhouse can be predicted by the vibration data of units and pressure pulsation data of the draft tube. The results show that the prediction model has high accuracy.