It is a very common phenomenon that airline passengers always instinctively choose the nearest exit to evacuate in emergency situations. However, such a decision may not be the wisest choice as airline passengers oftentimes lack awareness of aircraft configurations and evacuation capacity of each available exit. To improve the efficiency of aircraft emergency evacuation, a direction guidance system that guides passengers to escape from the optimal exits is proposed. Determining the optimal evacuation directions via the evacuation simulation model developed in our earlier work can essentially be viewed as an optimization problem. The kriging model, acting as a metamodel, is utilized to alleviate computational burden of running the simulation model and approximate the underlying relationship between inputs (evacuation signs) and outputs (evacuation time). To improve the accuracy of the kriging model at the region of interest, a sequential sampling strategy is used to add extra samples as a complement to the initial sample set to tune the initial kriging model. The decision validation metrics along with a termination criterion are used to validate the globally optimal solution with a certain degree of confidence. As demonstrated in the case studies, the proposed metamodel-based optimization method is computationally efficient and the evacuation time can be significantly saved.