The clustering in time and periodicity of earthquake occurrence are investigated statistically by the use of historical data of strong earthquakes in Tokyo. In order to discuss the periodicity of main shocks, the author examines the homogeneity of historical materials and the distributions of the intervals between any two earthquakes and between two successive earthquakes. It is found that one third to one fourth of events can be regarded as fore- and aftershocks in a wide sense. By the use of Monte Calro method, it is tested statistically whether the occurrence of main shocks is represented by a stationary random process or a periodic process. As a result, the 69-year periodicity, which is the highest peak in the periodgram calculated from Kawasumi's table, is not statistically significant at a 95% confidence level, but becomes significant if we lower the confidence level down to 80%. From the table of Usami and Hisamoto, the 36-year periodicity, which is the most predominant in their table, is found to be insignificant even at the 80% confidence level.