This paper reports the results of an experiment in which subjects had to buy real products in 5 different budget/price situations. Subjects were randomly drawn from the population of a medium-sized French city, and some of their socio-economic characteristics were recorded. We check the consistency of product choices with the Generalized Axiom of Revealed Preferences (GARP), and find that 29% of the subjects are GARP-inconsistent. This inconsistency rate is slightly lower than the ones found in comparable studies. A possible explanation for our lower rate of GARP violations might be that the subjects in our study were confronted with less budget/price situations. In looking for determinants of the GARP violations, we find that gender, the size of the household, the degree of switching between different products, and the times spent on performing experimental tasks have significant impacts on the probability of GARP-inconsistency.