Sensing mechanisms that estimate the occupancy of wireless spectrum are crucial to the success of approaches based on Dynamic Spectrum Access. In this paper, we present key insights into this problem by empirically investigating the design of sensing mechanisms applied to check the availability of excess capacity in CDMA voice networks. We focus on power-based sensing mechanisms since they are arguably the easiest and the most cost-effective. Our insights are developed using a unique dataset consisting of sensed power measurements in the band of a CDMA network operator as well as “ground-truth” information about primary users based on operator data. We find that although power at a single sensor is too noisy to help us accurately estimate unused capacity, there are well-defined signatures of call arrival and termination events. Using these signatures, we show that we can derive lower bound estimates of unused capacity that are both useful (non-zero) and conservative (never exceed the true value). We also use a combination of measurement data and analysis to deduce that multiple sensors are likely to be quite effective in eliminating the inaccuracies of single-sensor estimates.