Previous studies have indicated that data compression in wireless sensor networks is not always beneficial to energy conservation due to the additional computational energy costs. This work gives an energy-efficient arbitration mechanism that enhances the performance of compression algorithms by avoiding unnecessary energy losses. The adaptive compression arbitration system uses a new prediction modeling and adaptation. Tests show that the modeling method gives better predictions with the adaptive mechanism needing less data for modeling, which allows on-line prediction modeling instead of an off-line training process. Moreover, adaptive adjustments of the model parameters enable the system to keep good arbitration precision even as the data characteristics or applications change.