This work proposes a framework for jamming wireless networks that incorporates probabilistic models of internal states and observable characteristics of link protocols, where protocols are divided into two general classes: random access (RA) or channelized access (CA). Without exact knowledge of network parameters and internal state, the proposed intelligent jammer optimizes its strategy to be energy efficient while achieving the target throughput. Probabilistic models for jamming FDMA and CSMA-based protocols are described for illustration of the framework: A frequency-hopping voice network is analyzed to determine the optimal jam strategy for proactive frequency jammer; and a CSMA packet protocol is analyzed for varying packet arrival rates at the nodes. Since RA protocols display observable reaction to channel conditions, we propose a feedback-control loop that uses observable feedback to infer network parameters. Both protocols are evaluated through simulation for their energy-throughput tradeoff compared to a naive jammer.