Despite the large number of protein structures, we know little about their assembly into multi-protein complexes in cells. Such molecular machines are often transient, depend on their cellular environment, and therefore too complex, rare, and fragile to be purified, and therefore unsuitable for most structural techniques. I will illustrate the potential of EM tomography of cells and tissues with an emphasis on hearing machinery. I will describe the difficulties encountered, such as noise, macromolecular crowding and the complexity of the 3D data. I show how we currently overcome these challenges in order to achieve biological interpretations of the 3D cellular sceneries. We are using a variety of image processing analysis tools, mostly provided by others, such as