For the existing network entity geolocation algorithms based on the delay measurement are usually suffer from the inflation and jitter of the network delay, a new algorithm of city-level geolocation based on landmark clustering is proposed in this paper. According to the feature of the delays tend to be similar among the neighboring network entities generally, city-level network entities geolocation algorithm based on landmark clustering is given. In the proposed algorithm, the city-level landmarks are divided into several clusters based on the "delay similarity" principle to filters out the high-reliability landmarks for improving the precision of city-level geolocation algorithm, the minimal relative delay between cluster and target is calculated based on the minimal relative delay, and the region of this cluster is taken as the estimation of target's location. The rationality of the delay clustering, feasibility of the proposed algorithm, and experimental results show that the proposed algorithm reduces effectively the delay influence on the IP geolocation algorithms, and the geolocation accuracy is higher than the classic GeoPing algorithm.