A novel algorithm is proposed for the distributed and location-free detection of boundaries in sensor networks. The approach allows a node to decide autonomously, based solely on connectivity information of a small 2-hop neighborhood, whether it is in the interior of the network or on its fringes. This makes the presented algorithm well suited for scenarios that include mobility or dynamic changes to the network topology. The algorithm is compared qualitatively and quantitatively to multiple previous approaches. Various models and network settings are considered in extensive simulations. Even though the algorithm uses less information than most other approaches, it yields significantly better results. It is very robust against variations in node degree and does not rely on simplified assumptions of the communication model. Moreover, the approach is easy to implement on real sensor nodes, as it requires little computational power.