Abstract
In development of methods for functional dependencies recovery an approach is proposed. This approach allows to approximate functions, taking into account a number of factors that could significantly complicate and in some cases completely eliminate the possibility of using the classical statistical methods of bservations processing.