The evolution of mortality shows at older ages a marked deceleration. This phenomenon is generally thought to be an effect of selection: mortality decelerates because it progressively selects the most robust individuals in the cohort. Other processes, however, may contribute to mortality deceleration as well. People may not be passive in the face of ageing and may try to counter it by modifying their behaviours and lifestyles. In this paper, I propose a method to test whether selection is to be considered the unique responsible for mortality deceleration. I applied this method to the life tables of selected female cohorts drawn from the Human Mortality Database. The results indicate that between 75 and 90-95 years of age, mortality decelerates more rapidly than predicted by the selection theory. This rather simple observation challenges one of the central assumption in selection theory, the assumption that individual frailty is constant across life.