Generally， fault diagnosis based on vibration frequency analysis is used for high running reliability of the hydropower generating unit（HGU）. However， it is difficult to diagnose a fault correctly with only the spectrum feature because there is the same spectrum feature between different faults. In this paper， spatiotemporal characteristics in starting up a HGU are used in fault diagnosis by information fusion. The support vector machine（SVM） is adopted as information fusion algorithm at the feature level and D-S evidence theory is proposed as information fusion tool at the decision level. The experiment result shows that feature information is increased and reliability of fault diagnosis is improved by information fusion.