In October of 1993, there was decreased egg production and increased mortality among Mexican chickens, in association with serologic evidence of an H5N2 influenza virus. First isolated from chickens in May of 1994, after spreading widely in the country, the virus caused only a mild respiratory syndrome in specific pathogen-free chickens. Because eradication of the virus by destruction of infected birds posed major obstacles to the poultry industry in Mexico, we were able to conduct a "field experiment" to determine the fate of an avirulent virus after repeated cycles of replication in millions of chickens. By the end of 1994, the virus had mutated to contain a highly cleavable hemagglutinin (HA), but remained only mildly pathogenic in chickens. Within months, however, it had become lethal in poultry. Nucleotide sequence analysis of the HA cleavage site of the original avirulent strain revealed R-E-T-R, typical of avirulent viruses and unlike the K-K-K-R sequence characterizing viruses responsible for the 1983 outbreak in poultry in the United States. Both mildly and highly pathogenic isolates contained insertions and a substitution of basic residues in the HA connecting peptide, R-K-R-K-T-R, which made the HA highly cleavable in trypsin-free chicken embryo fibroblasts. Phylogenetic analysis of the HA of H5 avian influenza viruses, including the Mexican isolates, indicated that the epidemic virus had originated from the introduction of a single virus of the North American lineage into Mexican chickens. This sequence of events demonstrates, apparently for the first time, the stepwise acquisition of virulence by an avian influenza virus in nature.