A recent renaissance in small-angle X-ray scattering (SAXS) made this technique a major tool for the low-resolution structural characterization of biological macromolecules in solution. The major limitation of existing methods for reconstructing 3D models from SAXS is imposed by the requirement of solute monodispersity. We present a novel approach that couples low-resolution 3D SAXS reconstruction with composition analysis of mixtures. The approach is applicable to polydisperse and difficult to purify systems, including weakly associated oligomers and transient complexes. Ab initio shape analysis is possible for symmetric homo-oligomers, whereas rigid body modeling is applied also to dissociating complexes when atomic structures of the individual subunits are available. In both approaches, the sample is considered as an equilibrium mixture of intact complexes/oligomers with their dissociation products or free subunits. The algorithms provide the 3D low-resolution model (for ab initio modeling, also the shape of the monomer) and the volume fractions of the bound and free state(s). The simultaneous fitting of multiple scattering data sets collected under different conditions allows one to restrain the modeling further. The possibilities of the approach are illustrated in simulated and experimental SAXS data from protein oligomers and multisubunit complexes including nucleoproteins. Using this approach, new structural insights are provided in the association behavior and conformations of estrogen-related receptors ERRα and ERRγ. The possibility of 3D modeling from the scattering by mixtures significantly widens the range of applicability of SAXS and opens novel avenues in the analysis of oligomeric mixtures and assembly/dissociation processes.