Gossip-based peer sampling protocols have been widely used as a building block for many large-scale distributed applications. However, Network Address Translation gateways (NATs) cause most existing gossiping protocols to break down, as nodes cannot establish direct connections to nodes behind NATs (private nodes). In addition, most of the existing NAT traversal algorithms for establishing connectivity to private nodes rely on third party servers running at a well-known, public IP addresses. In this paper, we present Gozar, a gossip-based peer sampling service that: (i) provides uniform random samples in the presence of NATs, and (ii) enables direct connectivity to sampled nodes using a fully distributed NAT traversal service, where connection messages require only a single hop to connect to private nodes. We show in simulation that Gozar preserves the randomness properties of a gossip-based peer sampling service. We show the robustness of Gozar when a large fraction of nodes reside behind NATs and also in catastrophic failure scenarios. For example, if 80% of nodes are behind NATs, and 80% of the nodes fail, more than 92% of the remaining nodes stay connected. In addition, we compare Gozar with existing NAT-friendly gossip-based peer sampling services, Nylon and ARRG. We show that Gozar is the only system that supports one-hop NAT traversal, and its overhead is roughly half of Nylon’s.