Introduction: Graphic warning labels have been shown to be more effective than text-only labels in increasing attention and perceived health risks, but most U.S. studies have involved single exposures in laboratory or Internet settings. Methods: We recruited a convenience sample (N = 202) of U.S. adult smokers from population subgroups with higher rates of smoking and smoking-related deaths who had participated in a larger survey about graphic warning labels. Participants were randomized to get 1 of 9 graphic + text labels or a text-only label. Research staff affixed a warning label sticker to participants' cigarette pack(s) at enrollment. Color graphic labels covered slightly more than the lower half of packs. Black and white labels of current U.S. text-only warnings covered the existing side warning to prompt attention to the label (i.e., attention control). Participants received extra stickers of the same label for subsequent packs, and completed 3 telephone interviews in 1 week. Results: Participants reported low avoidance (<34%) and consistent use of the stickers (91%). Smokers consistently paid more attention to graphic than text-only labels. Only 5 of the 9 graphic warning labels were significantly associated with greater thoughts of health risks. Thinking about quitting and stopping smoking did not differ by label. Qualitative data illustrated differences in the "stickiness," self-referencing, and counterarguments of graphic warning labels. Conclusions: U.S. smokers' reactions to graphic warning labels on their own packs were similar to other, more controlled studies. Qualitative findings underscore the need for warning labels that encourage self-referential processing without increasing defensive reactions.