Sensor nodes are often organized into clusters to facilitate many network operations such as data aggregation. Clustering has to be protected in hostile environments. Otherwise, an attacker can easily mislead a cluster-based application by attacking the clustering protocol. This paper proposes three techniques for resilient cluster formation. The simple neighbor validation provides a simple yet effective way to validate a sensor?s neighbors; the priority-based selection organizes clusters based on the sensor?s priority of being a cluster head; and the centralized detection further enhances the security by detecting misbehaving nodes. Another appealing benefit of this protocol is that a sensor node can make a clustering decision immediately once the neighborhood information is available. This further increases the difficulty of attacking the clustering protocol. The analysis also shows that the proposed protocol is efficient and effective in dealing with attacks.