In order to effectively reduce frequency spectrum fragmentation and blocking probability in elastic optical networks, a spectrum allocation algorithm based on graph coloring model is proposed. Firstly, a graph coloring model is established for elastic optical networks, and the spectrum allocation problem is transformed into a graph coloring problem with weights. Then, two improved coloring algorithms(chain search method and improved greedy algorithm) are proposed. The simulation results show that compared with the degree maximum coloring algorithm, the two improved coloring algorithms can better optimize the spectrum resources and reduce the average blocking probability.