This paper aims to analyze how to determine the optimal concession period of BOT contract under demand uncertainty with the information about historical traffic demand. The previous literature about BOT contracts under demand uncertainty assumes that it is partially or fully resolved as soon as the highway is built, which may cause large estimation errors and lead to undesirable renegotiation for overlooking the demand fluctuation in the operation stage. We assumed that the growth rate of traffic demand is lognormal distributed and proposed an adaptive and dynamic prediction mechanism to predict the future traffic demand on the basis of statistical characteristics derived from the historical traffic demand, and then a decision model based on the NPV method was proposed to determine the optimal concession period with the average value and expect value of predicted traffic demand respectively. We compared the numerical results of the two types of concession periods and found that demand fluctuation in the operation stage did exist. It is more reasonable to consider the optimal BOT contract design with the expected value of future traffic demand because the risk of project failure can be quantified with probability.