Network size is the fundamental information of the distributed applications. Network size estimation methods must feature both high accuracy and adequate robustness in order to adapt to a large environment with a high node churn. Considering the fact that the existing network size estimation methods mainly focus on single optimization objective and fail to ensure accuracy and robustness simultaneously, a network size estimation method based semantic attraction—SEBSA is proposed in this paper. As the semantic information in SEBSA, hash values are hashed in real intervals by the peers’ identifies. The peers with adjacent hash values in SEBSA periodically exchange hash neighbors to attract the most adjacent peers in a hash space quickly. Meanwhile, every peer computes the average spacing among hash values of the hash neighbors to estimate network size. Theoretic analysis and experimental results reveal that compared with existing size estimation methods, SEBSA can provide accurate size estimation information quickly even in continually fluctuating network environment.