We developed a method for multiple alignment of protein sequences. The main feature of this method is that it takes the evolutionary relationships of the proteins in question into account repeatedly for execution, until the relationships and alignment results are in agreement. We then applied this method to the data of the international DNA sequence databases, which are the most comprehensive and updated DNA databases in the world, in order to estimate the 'evolutionary motif' by extensive use of a supercomputer. Though a few problems needed to be solved, we could estimate the length of the motifs in the range of 20 to 200 amino acids, with about 60 the most frequent length. We then discussed their biological and structural significance. We believe that we are now in a position to analyze DNA and protein not only in vivo and in vitro but also in silico.