Life history theory hypothesises that breeding events induce reproductive costs that may vary among individuals. However, the growing number of studies addressing this question are taxonomically biased, therefore impeding the generalisation of this hypothesis, especially with regard to marine top predators. This study investigated age-related survival and breeding performances in subantarctic fur seal (Arctocephalus tropicalis) females from Amsterdam Island, southern Indian Ocean. Using multistate capture–recapture models on data obtained from known-age tagged females over eight consecutive years, we tested for evidence of senescence, individual quality, and reproductive costs in terms of future survival and fecundity. Adult female yearly survival appeared high and constant throughout time. While a two age-class model was preferred in non-breeders, breeding females exhibited three age classes with a maximum survival for the prime-age class (7–12 years). Survival and reproductive probabilities decreased from 13 years onward, suggesting senescence in this population. Survival was lower for non-breeders than for breeders, among both prime-aged (0.938 vs 0.982) and older (0.676 vs 0.855) females. Furthermore, non-breeders exhibited higher probabilities of being non-breeders the following year than did breeders (0.555 vs 0.414). Such results suggest consistency in female breeding performance over years, supporting the hypothesis that non-breeding tend to occur among lower quality individuals rather than representing an alternative strategy to enhance residual reproductive value. However, the high proportion of females that did not breed during two consecutive years, and the lower probability of being a successful breeder after a greater reproductive effort confirmed the existence of reproductive costs, especially during the second half of the lactation. These results also suggest that younger age-classes included a higher proportion of lower quality individuals, which are likely to face higher costs of reproduction. Such hypotheses lead to consider the first breeding event as a filter generating a within-cohort selection process in females.