This work is a physical approach to eukaryotic cellular organization. Theses cells contains a specific distribution of filaments that enables them to change their shape, to move and to divide. These filaments and the associated proteins form the cytoskeleton, and are to the cell the equivalent of bones and muscles. The cytoskeleton self-organize to form caracteristic patterns that are universal in the eucaryotic family. We study the properties of self-organization on a model made from only one kind of filament. Our system is experimental, and is an invitro mixture of three proteins. The first one is tubulin, which constitutes micro-filaments called microtubules. The second one is kinesin, a motor protein able to bind a microtubule, and actively move on it's surface, using ATP as an energy source. The last protein, streptavidin, produces a complex of four kinesins. This small complex is able to physically connect two microtubules, and since it's made of motors, to move on both filaments. Due to the intrinsic polarity of the microtubules, all the complexes on a single filament move in the same direction, which leads to cooperative behavior and self-organization on the scale of the microtubule's length. We explored different geometries and bio-chemical conditions and obtained a few different patterns. In particular it can create asters, a radial distribution of microtubules. An aster formed by kinesin complexes is of inverse polarity compared to asters found invivo, but the process of self-organization by which they occur is universal, as can show a theoretical approach. We present some theoretical arguments that can predict the movement of filaments in specific geometries. We also compute an effect of confinement of the complexes in the geometry of an aster. We then describe a computer simulation based on molecular dynamics. Thanks to the very good characterization of the three proteins, the simulation is accurate, and can reproduce most of the patterns observed experimentally. We also use the simulation to explore new systems, made with more that one type of motor, and make predictions on the corresponding patterns. In doing so, we show that the simulation can be a precious help to experiments.