Background There is no evidence supporting that temperature changes COVID-19 transmission. Methods We collected the cumulative number of confirmed cases of all cities and regions affected by COVID-19 in the world from January 20 to February 4, 2020, and calculated the daily means of the average, minimum and maximum temperatures in January. Then, restricted cubic spline function and generalized linear mixture model were used to analyze the relationships. Results There were in total 24,232 confirmed cases in China and 26 overseas countries. In total, 16,480 cases (68.01%) were from Hubei Province. The lgN rose as the average temperature went up to a peak of 8.72℃ and then slowly declined. The apexes of the minimum temperature and the maximum temperature were 6.70℃ and 12.42℃ respectively. The curves shared similar shapes. Under the circumstance of lower temperature, every 1℃ increase in average, minimum and maximum temperatures led to an increase of the cumulative number of cases by 0.83, 0.82 and 0.83 respectively. In the single-factor model of the higher-temperature group, every 1℃ increase in the minimum temperature led to a decrease of the cumulative number of cases by 0.86. Conclusion The study found that, to certain extent, temperature could significant change COVID-19 transmission, and there might be a best temperature for the viral transmission, which may partly explain why it first broke out in Wuhan. It is suggested that countries and regions with a lower temperature in the world adopt the strictest control measures to prevent future reversal.