(eng): For applications to practical problems, the paper proposes to use the approximations of belief functions, which simplify their dependence structure. Using an analogy with probability distributions, we represent these approximations in the form of compositional models. As no theoretical apparatus similar to probabilistic information theory exists for belief functions, the problems arise not only in connection with the design of algorithms seeking the optimal approximations but even in connection with a criterion comparing two different approximations. With this respect, the application of the analogy with probability theory fails. Therefore, the paper suggests the employment of simple heuristics easily applicable to real-life problems.