A new method of contextual analysis of polynucleotide sequences in developed. The method finds nonrandom repeats in the sequences of N bases in length with given nucleotide frequencies. The coding regions of genes specifying beta-, beta'-, sigma-subunits of E. coli RNA polymerase were analyzed. The high content of short repeats was found to correspond to the secondary structure of globular proteins coded by the genes. The possible evolutionary role of the nonrandom direct repeats in coding regions of genes is discussed.