A modern program package for optimization methods generally uses schemes which, depending on certain parameters, are capable of generating a whole family of algorithms for the solution of a particular class of optimization problems. Given a problem in the package's subject area to solve, the task of selecting the best algorithm for the problem is quite complex. The authors attempt to develop an algorithm for the automated choice of the best algorithm in a program package for integer linear programming problems. They consider the algorithm selection problem in the following form: Let \(M=\{M_ 1,...,M_ l\}\) be a set of algorithms capable of solving the integer linear programming problems \(z_ 1,...,z_ q\), the descriptions and the optimal solutions of these problem being known. Let z be a new submitted problem whose description is known. Then select an algorithm from the set M which is the best by some given criterion for the solution of the problem z. The apparatus of classification theory [the first author and \textit{V. V. Nikiforov}, ibid. 3, 1-11 (1971; Zbl 0259.68042)] is used to develop the algorithm. No computational results are given.