The sixth national program for prediction of volcanic eruptron has been conducting since 1999 in order to advance toward the higher stage of the prediction of volcanic eruption. The progress of prediction depends not only on the advance of basic knowledge of volcanology but also on the accumulation of successful and non-successful practical knowledge through the occurrence of volcanic eruptions. Among nine eruptions and three volcanic unrests that occurred during the last five years in 1997-200, the predictions of two large eruptions at Usu and Miyakejima volcanoes in 2000 and of volcanic unrest at lwate volcano in 1998-1999 were reviewed and evaluated mainly on the basis of geophysical point of view. Based on detection of some precursory phenomena, the prediction of the beginning time of the eruptions at Usu and Miyakejima volcanoes was successfully performed by the practical experience. The semi-real-time prediction of the beginning time at Usu volcano was also successfully operated, based on the application of rock fracture criterion to the seismic data that were automatically processed in the real-time operation. However, it became clear that the predictions of eruptive mode and sequence at Miyakejima volcano were still far beyond the grasp of the basics because of abnormally complicated behavior. It is confirmed that a tomographic imaging of the shallow volcanic conduit system together with a precise 3-D hypocenter distribution contributes significantly for deep understanding of what was going on the volcanic unrest at lwate volcano.