A popular solution to dealing with large-scale social networks is to derive a representative sample from a social network. This sample is expected to represent the original social network well such that the sampled network can be used for simulations and analysis. In this paper, the authors propose a new social network sampling algorithm based on the temperature conduction model. Their sampling approach is able to effectively maintain the topological similarity between the sampled network and its original network. They have evaluated their algorithm on several well-known data sets.