In this paper, a new algorithm for dynamic airspace configuration is developed based on graph theory. A graph model is first constructed that accurately represents the air-route structure and air traffic in the National Airspace System. The airspace configuration problem is then formulated as a graph-partitioning problem to balance the subgraph (sector) workload while satisfying the capacity constraint, which is efficiently solved by a spectral clustering method. Since the original spectral clustering method shows some undesirable properties, such as disconnected subgraphs and unbalanced partitions, an algorithm is proposed to refine the partitions. Lastly, using the partitioned graph as an input, a novel airspace sectorization algorithm is developed, based on the graph search method. The new sectors computed by the sectorization algorithm have smooth boundaries and good geometrical shapes, which satisfy the minimum distance requirement (i.e., the sector boundaries are at least a minimum distance away from the airports, waypoints, and main air routes). The performance of the proposed dynamic airspace configuration algorithm is validated through various air-traffic scenarios with real air-traffic data.