Data matter. Protest data are no exception. Not all datasets areequal, nor do all have equal validity and reliability. Thus far,there has been no serious debate regarding the quality of protestdatasets. Increasing attention to protest studies suggests that nowis the time to talk about the sine qua non forgrowing knowledge in this sub-field of comparative politics. In thispaper I emphasize the importance of the underlying data sources. Inparticular, I argue that protest datasets should be drawn from manysources, particularly as many local sources as possible, in order toprovide accurate and meaningful data. To support this contention, Ifirst discuss traditional protest datasets and their limitations andprovide suggestions about how to overcome these limitations. Second,I explain an alternative type of dataset coding and itscharacteristics. And finally, I use statistical tests to illustratethe importance of local sources.I givemy sincere and deep thanks to Erik S. Herron and Ronald A.Francisco for reviewing drafts of this paper and giving meindispensable comments and continuous encouragements. I wouldalso like to thank Elizabeth Collins and Omur Yilmaz who editeddrafts of this manuscript. My appreciation also goes toanonymous reviews of this paper and the editors of PS:Political Science and Politics. I would like tothank my late farther, Ki-Dae whose love and support were simplyendless. I'll share any compliment with them, but only I willtake responsibility for any error.