Abstract. In this paper we present a new approach to power modeling and runtime power estimation for wireless network interface cards (WNICs). We obtain run-time power estimates by putting together four kinds of information: the nominal behavior of the card (taken from protocol and product specifications), its inherent power-performance properties (automatically characterized once for each card from digitalized current waveforms), the working conditions (characterized on the field by means of simple synthetic benchmarks) and the workload (observed at run time from the actual device). Experimental results show that our model provides run-time energy estimates within 3% from measurements.