Administrative data are considered the "gold standard" when measuring program participation, but little evidence exists on the potential problems with administrative records or their implications for econometric estimates. We explore issues with administrative data using the FoodAPS, a unique dataset that contains two different administrative measures of Supplemental Nutrition Assistance Pro-gram (SNAP) participation as well as a survey-based measure. We first document substantial ambiguity in the two administrative participation variables and show that they disagree with each other almost as often as they disagree with self-reported participation. Estimated participation and misreporting rates can be meaningfully sensitive to choices made to resolve this ambiguity and disagreement. We then document similar sensitivity in regression estimates of the associations between SNAP and food insecurity, obesity, and the Healthy Eating Index. These results serve as a cautionary tale about uncritically relying on linked administrative records when conducting program evaluation research.