For many years, the increasing development of microcomponents based products requires more and more microassembly systems to permit the assembly of micrometric parts and MEMS. These systems require different tasks such as micromanipulation, microassembly, supervision, quality inspection and also require accurate metric measurements from images. The most widely used image sensor is the photon videomicroscope which is characterized by several weaknesses such as the depthof- field and the field-of-view. This paper deals with the calibration of this kind of imaging system used for robotic microassembly. In the first part, a multiscale calibration paradigm is proposed and is used to pick and place silicon microparts ( 400×400×100μm3 ) using multiscale 2D visual servoing. In a second part, these microparts are assembled by a 3D visual control approach. High precision, repeatable results are obtained during the MEMS manipulation and assembly tasks.