Can a smartphone learn our eating habits without the user being in the loop? Clearly, the phone could use checkins based on location to infer that if you were in a cafe, for example, there is a good possibility you might eat or drink something. In this paper, we use inferred behavioral data and location history to predict if you are going to eat or not in the near future. These predictors could serve as a basis for future eating trackers that work unobtrusively in the background of your phone rather than relying on burdensome user input. In this paper, we report on a simple model that predicts the food purchases of a group of undergraduate college students (N=25) using inferred behavioral and location data from smartphones. The 10-week study uses the dining related purchase records from student college cards as ground-truth to validate our prediction model. Initial results show that we can predict food and drink purchases with an accuracy of 74% using three weeks of training data.