Tax evasion is an illegal concealment of a taxable activity and a worrisome phenomenon. Understanding how to motivate individuals to pay their taxes has become one of the main targets for policy makers. From the standard economic literature we know that a taxpayer makes a decision under risk, maximizing his expected utility. In particular, the payment of taxes is a highly structured decision process. Indeed, by choosing whether to comply with the law or to engage in the tax evasion, individuals take into account a set of social norms along with their role in the society. The behavior of other individuals within the same community matters since it increases the moral cost of non-compliance. But how does it affect individual decisions? Given the magnitude of this phenomenon and its economic consequences, a better understanding of its determinants represents an essential starting point for the definition of effective policy interventions. By using an experimental approach, in this paper we study how the type of information provided to our individuals within the same network affects their level of tax compliance. Social information has a particularly strong impact on tax compliance, especially when information flows within an interconnected network of individuals. The type of information subjects are provided with reveals to be pivotal in their decision process, especially when it refers to subjects belonging to their own reference group. This is because individuals attribute a greater value to the behavior of people they know better. To this end, by introducing the same monetary incentives to cheat, in our experiment design we test four different treatments: No Info, Full Info, PositiveInfo and NegativeInfo. In our baseline treatment, NoInfo, individuals are provided with full information on the event to be audited, the relative outcome, and their final payoff. Then, by manipulating the type of information given to individuals on their neighbors’ behavior, we explore the other three types of treatment. Specifically, in the PositiveInfo treatment, subjects receive information on the neighbors they are connected with, by obtaining details on the audit event, and its relative output, i.e. if their neighbors have been found complaint with the tax payment of the requested taxes. In the NegativeInfo treatment, instead, subjects are again informed about their neighbors about the audit event, but they will be told only if their neighbors have been caught non-compliant. Finally, in our last treatment FullInfo, participants receive complete details on the others two 2 neighbors there are connected with, which means subjects will be fully aware of the audit event, and its relative output either when their own neighbors have been complainant, or not.