Based upon their previous studies in the concentration of deaths in winter in Japan and the West European countries, on the one hand, and, on the other hand the considerable moderation of seasonal variation in mortality, or rather “ deseasonalization ” of deaths, in the Scandinavian countries and the United States in this paper the authoresess attempt to analyze quantitatively the difference between these two types of seasonal variation in mortality and specifically the geographical difference in relationship between mortality (stroke and infant) and temperature in the three cities of London, Tokyo and New York, and they sincerely want to build up one of the groundworks necessary for long-term biometerorological forecast. For such statistical analysis, census method II and covariance analysis in the twoway layout are used.1. As for the seasonal variation of mortality, time series analysis appears to be the effective statistical methods. Thus, census method II has been adopted to analyze a time series of stroke mortality. Generally, the time series is composed of the trend cycle (TC) the seasonal variationcomponent (S) and the irregular component (I), but the second named alone is studied in this paper because it calls for serious considerations from the biometeorological standpoint. The time series cycle of the seasonal variation component is naturally regular in the three cities, but the amplitude is much smaller in New York than in London and Tokyo. The correlation coefficient between stroke mortality and temperature in terms of the seasonalcomponent comes at -0.9709 for Tokyo, or slightly larger than -0.9281 for London and -0.9015 for New York. As there is some time lag between mortality and temperature, the correlation coefficients are calculated with one to five months shifted, and it is seen that the coefficient is the largest for Tokyo in case there is no time lag, and that it appears to reach the peak, if half a month is shifted, for London and New York. Details of the time series analysiswith census method II will be published later.2. Covariance analysis indicates that there is no significant difference among the three cities between the regression equation of mortality (stroke and infant) on temperature in spring and that in autumn in the 1960's, so the changes per 10°C in the death rate are calculated by average regression coefficient. Mortality is inversely related to temperature, and the increase or decrease per 10°C in the death rate for stroke turns out to be larger than that for infant mortality in the three cities. In both Tokyo and London, the rise of the death rate for stroke in autumn is considerably large, or 24.0% and 32.4%, respectively, contrasted to only 8.3% in New York (over three and four times larger, respectively).As for infants, the mortality rate decreases or increases very slightly (not more than 1.5%) in New York along with the ups and downs of temperature. The corresponding rate of increase or decrease comes at about 8-9% in London. In Tokyo, however, the infant death rate declines by about 16.0% with tempearture up 10°C and goes up by as much as 17.5% with the similar drop of temperature.The changes in death rate with the variation of temperature turn out to be the smallest in New York among the three municipalities. All this clearly indicates that central heating, widely adopted in the American city, has brought about favorable bearings upon mortality, especially that of infants who can live in the artificial climate. It is also to be noted that mortality from stroke undergoes much larger changes with the ups and downs of temperature in London than in Tokyo, and vice versa in the case of infant mortality.