The study and use of sparse representations in data-rich applications has garnered signicant attention in the signal processing, statistics, and machine learning communities. In the present work we describe a novel sensing procedure called Distilled Sensing (DS), which is a sequential and adaptive approach for recovering sparse signals in noise. Passive sensing approaches, currently the most widespread data collection methods, involve non- adaptive data collection procedures that are completely specied before any data is observed. In contrast, DS collects data in a sequential and adaptive manner. Often such procedures are known as active sensing or sequential experimental design, and allow the use of data observed in earlier stages to guide the collection of future data. The added exibility of active sensing, together with a sparsity assumption, has the potential to enable extremely effcient and accurate inference.