Data on centenarian deaths by detailed age have been brought together from 17 countries. After testing, the data for 13 of them have been judged basically reliable for analysis, though with some corrections. This material which comprises about 145,600 person-years-at-risk at 61,200 deaths has been analysed by the method of extinct generations, extending it partly to living cohorts with registration data or with estimation. In all reliably documented populations the risk of dying continues to rise with advancing age for both sexes probably tending to an asymptote. The risk for men consistently exceeds that for women. The annual probability of dying is of the order of 0.75–0.80 at about age 108 among men and 110 among women. Survivors are rare at such ages, and only exceptionally have individuals been reliably shown to have exceeded these ages by a few years. In many countries the mortality of centenarians has been declining over past several decades, particularly among women and among persons below the age 105. The theory that natural selection through death leaves survivors with lower-than-average mortality, is disproved by empirical data which show a slight tendency to the contrary. The explanation offered is that frailty is not constant, but influenced by late effects of temporary risk factors. While the survival of centenarians is improving at least in some countries, itis not possible to determine whether the natural life span has increased. The upper end of the human life span would be better understood if old age were accepted for classification either as the underlying, or as a contributing cause of death when so diagnosed.