We have analyzed solar eclipse records taken from Japanese history books covering the period from 628 to 1800. A celestial-mechanical computation shows that among 685 eclipse-related records 370 cases correspond to eclipses actually occurred in the Northeast Asia. Seventy-two percent of the records appearing from 950 to 1550 correspond to the eclipses that can be observed in Japan. But only 35% of eclipse records before 950 correspond to observable eclipses. The ratio is also small(47%) during the period between 1550 and 1650. In order to study the collective properties of eclipses recorded in Japanese history books we grouped the eclipse records in such a way that each group contains 30 confirmed consecutive records. Then for each group 30 eclipse magnitude maps are averaged to measure the degree of concentration of moon's shadow and to seek for the location of concentration where the group of 30 eclipses can be optimally observed. We have found that the degree of concentration of moon's shadow is high for the eclipses from 1000 to 1550 and after 1650, and that the location of concentration is close to Japan. In particular, the location is exactly at Japan during the Kamakura (鎌倉) period. However, for eclipses recorded before 950 and around 1600 the degree of concentration is very low and the optimum observing location is far from Japan. This tells that the majority of eclipse records during these periods are not based on real observation. Our analyses consistently show that the properties of Japanese eclipse records are tightly related with political changes. We have also noticed an exceptional period. The eclipses recorded during the Yamato (大和) period before 710 are found to concentrate on a place near the South China Sea, more than 20° away from Japan. A set of simulations indicates that these eclipses have collective properties incompatible with those recorded by Japanese at later periods.