Bertrand supergames with non-binding communication are used to study price formation and stability of collusive agreements on experimental duopoly markets. The experimental design consists of three treatments with different costs of communication: zero-cost, low-cost and high-cost. Prices are found to be significantly higher when communication is costly. Moreover, costly communication decreases the number of messages, but more importantly, it enhances the stability of collusive agreements. McCutcheon (1997) presents an interesting application to antitrust policy by letting the cost of communication symbolize the presence of an antitrust law that prohibits firms from discussing prices. Although our experimental results do not support the mechanism of McCutcheon's (1997) argument, the findings point in the direction of her prediction that antitrust laws might work in the interest of firms.