The article raises the problem of the interpretation of economic data, including macroeconomic ones, emphasizing the fact that these data are often contradictory. To solve this kind of economic problems, the authors consider a neural network. First of all, we analyze the simplest neural network with one neuron, which will serve as a separating function. Then, we consider the theorem of the existence of divided neural network (functions) and offer its proof. The transition to the infinite case, that is to the neural systems is substantiated. The conclusion of the non-formalizable some economic problems is drawn, and basic sources that reveal their methods of solution are given. One of the most important approaches is committee method.