Vital statistics data previously published by the Registrar General of England and Wales have been separated into male and female deaths in 1960-69 aged 100, 1961-70 aged 101, 1962-71 aged 102 etc., to represent cohorts of 292 male and 1463 female centenarians born in 1859-1869. Mortality is fitted to age by the method of maximum likelihood. Estimated male mortality declines after age 103 and falls below that of females (who show no decline) at age 105. The cause of this non-Gompertzian tapering off in mortality is found to be selection due to variability in the males, which is more than sufficient to offset the biological effects of aging that nevertheless occur in each individual. Selection causes a reduction of 0.01 per year of age in mean male centenarian mortality rates in England and Wales.