Bayesian networks have become one of the most popular multidimensional models for uncertain knowledge formalization. Possibilistic belief networks that are their possibilistic counterpart, though not so popular, seem to possess a couple of minor but pleasant advantages: the distributions are easier to modify to adopt expert subjective knowledge, and conditioning in possibility theory is dependent on a selected t-norm and therefore the apparatus seems to be more flexible.
The main part of this paper describes a new way of defining possibilistic belief networks as a sequence of low-dimensional distributions connected by operators of composition. To make this nonstandard technique more easily comprehensible, we introduce an extended motivation part explaining the basic notions through a simple (probabilistic) example. The paper is concluded by a lucid example: an application of the apparatus for fundamental analysis of engineering enterprises.