Résumé
Our article presents an experimental assessment of a tax-evasion prevention mechanism proposed by Greenberg (1984). Greenberg argued that tax evasion cannot totally be eliminated in a population. On the basis of this observation, he developed a mechanism in which the audit probability is determined by agents' earlier fraud behavior. We observe that the distribution of subjects differs from the model's prediction. The discrepancy is due to the existence of an increasing relationship between the decision to commit fraud and income--a relationship that the model does not take into account. Our results suggest that Greenberg's mechanism is based on an over-restrictive hypothesis.