The tumbleweed algorithm (TA) is a new optimization algorithm inspired by a plant called tumbleweed, and like other evolutionary algorithms, this algorithm is used to solve parameter tuning in real-world problems. The algorithm simulates the process of tumbleweed from young shoots to maturity as well as the process of sowing seeds after maturity, and a growth cycle formulation is added to combine these two processes. In this paper, the TA algorithm is applied to the multilevel image segmentation problem, solving the problem on the basis of minimum cross-entropy thresholding. The experimental results show that the TA algorithm has good image segmentation results.