Abstract: The distance prediction performances of landmark-based IP coordinate system are sensitive to the selection and placement of landmarks and the malicious behaviors of some hostile nodes, which have greatly impact on the accuracy and trustiness of the predicted distances for most of applications. This paper proposes a new landmark-based IP coordinate system with defend-capable malicious behaviors (for short LCSD), in which the selection and placement of landmarks are optimized through the distance matrix clustering scheme, the malicious behaviors of hostile nodes are prohibited to disturb the coordinate update of the normal nodes through the cooperative recommending trust-evaluation mechanism. Finally by some simulations, its performances are analyzed in terms of the relative error (RE), the neighbor approximation degree. The closest neighbor trust degree, and the results show that LCSD has much smaller RE on the whole, and outperforms ICS in the neighborhood in despite of the proportion of hostile nodes to the all nodes.