Abstract: In order to estimate energy overhead of sensor nodes, it is necessary to model their power consumption reasonably and accurately. However, existing power models are not accurate enough. This paper proposes a novel power model based on power state machines with continuous parameters, which can be used for any type of sensor nodes. This model is able to predict power consumption of sensor nodes according to changes of supply voltage and operating frequency. By handling ATmega128 (L) microprocessor frequently used in typical sensor nodes with this model and comparing with independent measurement results, it is shown that prediction error of this model is less than 1% for active state and 9.7% for idle state. This model can be used to substitute existing models of sensor network simulators to provide more accurate prediction results of energy overhead for sensor nodes.