We develop a new approach, which allows us to study structural properties and derive generalized characteristics for the tables of `object-object’ and `object-feature’ types. This method is based on metrical properties of entropy. There is no preliminary normalization of incommensurable data. In order to solve the problem of approximation of matrices by vectors, it is proposed to reduce this problem to searching for optimal strategies of infinite antagonistic game. To illustrate advantages of the methods developed, the examples of clasterization are used.