We have analyzed the popularization activities undertaken by ten thousand CNRS researchers by means of their annual reports for the years 2004, 2005 and 2006. This is the first time that such an extensive statistical study on science popularization practices is carried out. Our main findings are : - the majority of researchers is not involved in popularization (51% has not done any popularization over the three-year period, two thirds have been involved in no more than one popularization action). - popularization practices are extremely diverse, both at the individual level (we have identified three subpopulations that feature distinctive attitudes towards popularization), and at the level of scientific disciplines (researchers in Humanities are twice as active as the average), as well as in laboratories or geographical regions. - the number of actions reported in 2005 greatly increased compared to 2004 (+ 26%), while they slightly diminished in 2006.