We propose an agent centric algorithm that each agent (i.e., node) in a social network can use to estimate each of its neighbor’s degree. The knowledge about the degrees of neighboring nodes is useful for many existing algorithms in social networks studies. For example, algorithms to estimate the diffusion rate of information spread need such information. In many studies, either such degree information is assumed to be available or an overall probabilistic distribution of degrees of nodes is presumed. Furthermore, most of these existing algorithms facilitate a macro-level analysis assuming the entire network is available to the researcher although sampling may be required due to the size of the network. In this paper, we consider the case that the network topology is unknown to individual nodes and therefore each node must estimate the degrees of its neighbors. In estimating the degrees, the algorithm correlates observable activities of neighbors to Bernoulli trials and utilize a power-law distribution to infer unobservable activities. Our algorithm was able to estimate the neighbors’ degrees in 92% accuracy for the 60867 number of nodes. We evaluate the mean squared error of accuracy for the proposed algorithm on a real and a synthetic networks.