Low power is a primary concern in the field of wireless sensor networks. Bluetooth has often been labeled as an inappropriate technology in this field due to its high power consumption. However, most Bluetooth studies employ rather over?simplified, fully theoretical, or inadequate power models. We present a power model of Bluetooth including scatternet configurations and low?power sniff mode and validate it experimentally on a real Bluetooth module. Based on this model, we introduce a power optimization framework employing MILP (Mixed?Integer Linear Programming) techniques, and devise optimal power management policies in the presence of end?to-end delay constraints. Our optimizations, if backed by power?aggressive hardware implementations, can make Bluetooth viable for a wider range of sensor networks.