Sensor networks have posed challenges in programming, routing and security that are quite unique. With respect to security, there are a number of considerations that separate sensor networks from traditional systems, like resource constraints, possibilities of attack, and even aggregate nature of sensor data. These differences merit a rethinking of current approaches to security on several fronts such as tolerating compromise of a few nodes while preserving semantics of the application. The emphasis of our work is on developing mathematical models and techniques for protecting the sensor network from possibly new attacks by eavesdroppers and hackers.