Abstract : The stochastic nature of gene expression is now clearly established experimentally and appears as an essential part of the cellular dynamics. An important source of this variability comes from the dynamics of the different multiprotein structures involved in the process of gene expression. Using a modeling approach, we study here how the interactions between molecules which individual behavior is probabilistic can result in global dynamics and can influence the expression level. We focus more precisely on two aspects of the gene expression process: first, its spatial context within a structured and dynamic nuclear landscape, and, second, the combinatorial stochastic molecular events occurring on the promoter of a gene. For the study of mesoscopic organization phenomena in the cell nucleus, we propose a "4D" simulation model (considering space and time). It is built upon different techniques taken from higher and lower scale formalisms (ie. molecular simulations and cellular models), while keeping the essential aspects for our purpose (individuality of some molecules, steric volume exclusion, electromagnetic interactions, chemical reactions, . . . ). For investigating the stochastic dynamics of transcriptional regulation, we propose a second model describing the events of association/dissociation and chromatin modification, based on the cooperative/competitive affinity of molecules and their potential enzymatic or remodeling activity. Using analytical and computational techniques, we describe the activity of the promoter with tools from signal theory, but also by reproducing the measures of various experimental techniques (ChIP kinetics, FRAP, FRET, flow cytometry, . . . ). The analysis of the model reveals that the spontaneous activity of a promoter can be highly complex and demonstrate a multi-scale dynamics similar to what is observed experimentally (rapid turnover of molecules, slow cyclical behavior, transcriptional heterogeneity, . . . ). Finally, we show that confronting experimental data of various nature can reveal the structure of the underlying system. This model appears as a general theoretical framework for the investigation of promoter dynamics and the interpretation of experimental data.