The Bayesian Agent approaches the problem of incomplete information in closed negotiation by learning an opponent's preferences by studying the negotiation moves that an opponent makes during the negotiation. The learning technique is based on a Bayesian learning algorithm. The opponent model consists of a set of hypothesis about evaluation functions and weights of the issues. The probability distribution is defined over the set of hypothesis that represent agent's belief about opponent's preferences. The agreement reached by the Bayesian agents usually has a higher utility than that reached by the other strategies.