Based on vital statistics from 1979 to 1983, we used a time-series analysis using the Box-Jenkins model (ARIMA model) to compare the seasonality of preterm births among 47 prefectures in Japan. We also evaluated the relation between seasonality of preterm births and temperature. According to the Box-Jenkins analysis, a seasonal moving average was detected for 44 prefectures. The seasonality of preterm births in general shows a bimodal pattern with two peaks, one in summer and one in winter. The most interesting result is that the peaks show a trend with geographic location; one peak is dominant in winter among the northern prefectures and the other peak is dominant in summer among the southern prefectures. Another important result is that the winter increase in preterm births is negatively correlated with mean winter temperature (R = -0.424, p = 0.003); on the contrary, the summer increase in preterm births is positively correlated with the average summer temperature (R = 0.549, p < 0.001). These results suggest that climate-associated factors should be associated with the occurrence of preterm births in Japan.