All current urban models accept the 'first-order recursion' view, namely, that the state of an urban system at time t is sufficient for predicting its state at t + 1. This assumption is not at all evident in the case of urban development, where the behavior of developers and planners is defined by the complex interaction between long-term and short-term plan guidelines, local spatial and temporal conditions, and individual entrepreneurial activity and cognition. In this paper we validate the first-order recursion approach in an artificial game environment: thirty geography students were asked to construct a 'city' on the floor of a large room, with each student using the same set of fifty-two building mock-ups. Based on the analysis of game outcomes, the first-order recursive set of behavioral rules shared by all the participants is estimated and further employed for computer generation of artificial cities. Comparison between the human-built and model patterns reveals that the constructed set of rules is sufficient for representing the dynamics of the majority of experimental patterns; however, the behavior of some participants differs and we analyze these differences. We consider this experiment as a preliminary yet important step towards the adequate modeling of decision-making behavior among real developers and planners.