Six pairs of experimenters observed 1482 persons entering food and non-food locations during lunch hour and categorized them as obese, borderline, or normal weight. The two main hypothesis, that obese individuals would be over-represented at eating establishments in general and at those featuring high calorie dessert items in particular, were not supported. Nevertheless, two important methodological findings emerged. Observer expectations about the results of the study exerted a modest but systematic effect on the data. In addition, although interobserver reliability was high, there were marked differences between observer pairs in their criteria of obesity. Pretraining blind observers to meet objective criteria of adiposity is recommended for naturalistic assessments of obesity.