We have investigated amino acid substitutions in hemagglutinin HA1 of influenza viruses A/H1N1, A/H3N2, B and A/H5N1 in human patients from year to year. An important antigenic epitope may consist of two amino acid residues each from the 130- and 220-loops around the sialic acid binding site. Antibodies against that epitope should block the interaction between hemagglutinin and sialic acid.  The temporal changes of these four residues for A/H1N1 and A/H3N2 follow specific trends.  On the other hand, for B and A/H5N1, these residues remain invariant. Other amino acid residue changes are few for the same strain isolated within each year or nearby years, especially for those with same loop residues. It is therefore suggested that if the loop residues remain unchanged, the latest strain should be used for flu vaccine of the next season, as A/California/2004(H3N2) for 2005/2006. On the other hand, a more recent strain should probably be used for A/H1N1. If the loop residues are changing, two different isolates may be required.  For B, there are two isolates differing by about 30 residues yearly. However, they have changed little during the past decade.  Both of them should thus be used for vaccine but only once every ten years.  For A/H5N1, isolates in 2004 differed by only a few residues. One of these has been used to produce a flu vaccine strain. Together with drugs designed against neuraminidase and matrix protein, the phobia of A/H5N1 pandemic similar to the 1918 (A/H1N1) pandemic is probably unfounded.