Life‐table data of 56 natural populations of mammals were analysed, modelling mortality rate as a three‐phase step function of age (the step model) instead of using the Gompertz model. In the step model, mortality rale is constant in each phase. The phases correspond to juveniles and young and old adults. The age of transition between young and old adults is referred to as the age of senescence. This approach has the advantages that, for the first time, the age of senescence is identified objectively using a robust statistical procedure, and that young adult mortality rates are estimated without bias since no assumption is made about how adult mortality rate changes with age. A further statistical problem solved here that has previously caused difficulty is that of correctly accounting for the different levels of precision in data from different age classes.Significant changes (P < 0.05) in adult mortality rate with age were found in 27 out of 56 populations. In 23 of these 27 cases, adult mortality rate increased with age. Juvenile mortality rate differed significantly from young adult mortality rate in 21 cases; in 18 of these the rate for juveniles was higher. These results are discussed in relation to earlier analyses, in particular that of Promislow (1991).