The Walrasian general equilibrium model is the centrepiece of modern economic theory, but progress in understanding its dynamical properties has been meagre. This article shows that the instability of WalrasÕ tâtonnement process is due to the public nature of prices, which leads to excessive correlation in the behaviour of economic agents. When prices are private information, a dynamic with a globally stable stationary state obtains in economies that are unstable in the tâtonnment process. We provide an agent-based model of a multi-sector Walrasian economy with production and exchange, in which prices are private information. This economy is dynamically well behaved.In WalrasÕ original description of general equilibrium (Walras, 1954(Walras, [1874), market clearing was effected by a central authority. This authority, which has come to be known as the ÔauctioneerÕ, remains today because no one has succeeded in producing a plausible decentralised dynamic model of producers and consumers engaged in market interaction in which prices and quantities move towards market-clearing levels. Only under implausible assumptions can the continuous ÔauctioneerÕ dynamic be shown to be stable (Fisher, 1983), and in a discrete model, even these assumptions (gross substitutability, for instance) do not preclude instability and chaos in price movements (Saari, 1985;Bala and Majumdar, 1992). 1 Moreover, contemporary analysis of excess demand functions suggests that restrictions on preferences are unlikely to entail the stability of tâtonnement (Sonnenschein, 1972(Sonnenschein, , 1973Debreu, 1974;Kirman and Koch, 1986).It has been a half century since Debreu (1952) and Arrow and Debreu (1954) provided a satisfactory analysis of the equilibrium properties market economies, yet we know virtually nothing systematic about Walrasian dynamics. This suggests that we lack understanding of one or more fundamental properties of market exchange.This article provides an agent-based model of the Walrasian economy. An agentbased model is a computer simulation of the repeated play of a game in which a large number of agents are endowed with software-encoded strategies governing both how they play the game and how they gather information and update their behaviour. The disequilibrium behaviour of agents in our agent-based models is governed by a replicator dynamic (Taylor and Jonker, 1978) in which, over time, successful agents tend in Darwinian fashion to increase in frequency at the expense of unsuccessful agents. We describe the process of shifting from lower to higher payoff strategies as ÔimitationÕ, * I thank Kenneth Arrow, Robert Axtell, Samuel Bowles, John Geanakoplos, Peter Hammond, David Levine, Axel Leijonhufvud, Charles Plott, Sharon Harrison, Leigh Tesfatsion, two reviewers, and participants in seminars at the Santa Fe Institute, Columbia University, UCLA, and Trento for helpful comments, and the John D. and Catherine T. MacArthur Foundation for financial support. The computer programs used in this article are available from the author....