The trouble starts if a tandem duplication , occurs in S1 but only one instance of the motif, M2, is present in S2. Here, it can happen that the beginning of M2 is aligned to the beginning of and the end of M2 is aligned to the end of as in Figure 1 (B). DIALIGN is particularly susceptible to this type of errors since it does not use gap penalties. The situation is even more problematic for multiple alignment. Consider, for example, the three sequences S1, S1, S3 in Figure 1 (C), where two instances , of a motif occur in S1 while S2 and S3 each contain only one instance of the motif M2 and M3, respectively. Under the above assumptions, a biologically meaningful alignment of these sequences would certainly align S2 to S3, and both motifs would be aligned either to or to – depending on the degree of similarity of S2 and S3 to and , respectively. Note that such an alignment would also receive a high numerical score since it would involve three pairwise alignments of the conserved motif. However, since the pairwise alignments are carried out independently for each sequence pair, it may happen that the first instance of the motif in sequence S1, is aligned to M2 but the second instance, , is aligned to M3 in the respective pairwise alignments as in Figure 1 (C). Thus, the correct alignment of M2 and M3 will be inconsistent with the first two pairwise alignments. Depending on the degree of similarity among the motifs, alignment of and M3 may be rejected in the greedy algorithm, so these motifs may not be aligned in the resulting multiple alignment. It is easy to see that the resulting multiple alignment would not only be biologically questionable, but it would also obtain a numerically lower score as it would involve only two pairwise alignments of the motif.