Sensors are all around us, in buildings, vehicles and public places, from commodity thermostats to custom sensornets. Yet today these sensors are often disconnected from the world, either because they are distant from infrastructure, and wide-area networking (by 3G cellular, satellite, or other approaches) is too expensive to justify. Data muling makes communication cost-effective by leveraging short-range wireless and mobility, perhaps by zebras, buses or farmworkers. In this paper we propose that human-carried mobile phones can serve as data mules for sensornet deployments, exploiting ubiquity of mobile phones and human mobility to bring low-cost communication to sensors. We use two mobile phone datasets to show that Bluetooth can serve as a viable muling network, and humans already see many potential sensors regularly. We have implemented a mobile-phone-based data muling system, and used it in four sensornet deployments totaling ten months operation. We find that muling can be the only cost-effective option for rural deployments, where it is critical to monitoring remote sensor networks. We also show opportunistic mobility can collect data without any extra effort in residential and office environments. Finally, we systematically evaluate our deployments to understand how contact duration and data size interact, and to evaluate the effect of muling on phone batteries.