The global genetic optimization algorithm under consideration uses the normal distribution and has a number of significant advantages as compared with the random search algorithm with “memory”. The method in question is applied to the problem of beam dynamics optimization in a linear waveguide accelerator; the problem is reduced to global extremum search of quality functional in multidimensional region of control parameters variation. The algorithm quickly converges and can be successfully used in solving multidimensional optimization problems, including its combination with directed methods.