There has been so far no global estimate of snowfall. CloudSat has, for the first time, provided an opportunity for us to conduct such an estimate. The present study seizes this opportunity and attempts to investigate the global snowfall characteristics using its cloud radar observations. The retrieval methodology developed in this study includes two parts: first, determining whether a radar echo corresponds to snowfall (instead of rainfall), and second, converting radar reflectivity to snowfall rate. The first part is a snow‐rain threshold based on multiyear land station and shipboard present weather reports, and the second part is based on backscatter computations of nonspherical ice particles and in situ measured particle size distributions. Using the above retrieval method, global CloudSat data over 1 year were analyzed. The results show the following. (1) In the Southern Hemisphere, there is an almost zonally orientated high snowfall zone centered around 60°S, where both snowfall frequency and rate are high. In the Northern Hemisphere, however, heavy/frequent snowfall areas are mostly locked to geographical locations. (2) Zonally and annually averaged snowfall rate has its maximum value around 2 mm d−1, which is about one third of the zonally averaged rainfall values found in the tropics, signifying the importance of snowfall in hydrological cycle. (3) Vertical profiles of snowfall rate have the greatest variability in the lowest levels. While near‐surface snowfall rate generally increases with cloud top height, there seems to be two prevailing groups of clouds with very different growth rate of snowfall as cloud top height increases. (4) The characteristics of the vertical distribution of snowfall rate are quite similar for over‐ocean and over‐land snow clouds, except that over‐land snow clouds seem to be somewhat shallower than those over ocean.