We conducted a survey and a randomized natural experiment with the same subjects to investigate the effect of information about calorie intake on fast food choices. This combined approach allows us to maximize both internal and external research validity and test consistency of findings. We find that providing information about calories in a survey context for fast food menus has a moderate effect on calorie consumption, decreasing on average by 2.96 percent the amount of calories of the selected food choices. However, the same nutritional information had no significant impact on actual purchases in the restaurant context. Among the possible menus, the salad menu (the healthiest menu) was the most preferred option by those respondents who received nutritional information in the survey context; whereas in the restaurant, the most popular choice for the same group of people was the “Double bacon burger option” (the least healthy option). Finally, we find that the average calorie content of participants’ actual purchases increases significantly (0.17%) with the number of days elapsed between the day when the survey took place (and information was provided) and the actual purchase day at the restaurant. These results show large discrepancies between stated preferences and actual market behavior. These findings may be justified by the existence of projection bias and subjects acting under rational ignorance.