The molecular distinctions between mortality stages 1 (M1; senescence) and 2 (M2; crisis) of human replicative aging are ill defined. We demonstrate a qualitative difference between telomeric end associations at M1 and the end fusions that produce dicentric chromosomes and breakage-fusion cycles. Knockdown of ligase IV sufficient to completely inhibit radiation-induced dicentric chromosome formation had no effect on the frequency of telomere associations (TAs), establishing that TAs are not covalent conventional nonhomologous end-joining (NHEJ) products. TAs preceded and were more numerous than dicentric chromosomes. Cells initially tolerated dicentric chromosomes without dying, but eventually, a combination of too many TAs and dicentrics/complex chromosomal rearrangements resulted in apoptosis. We propose a working model in which end associations represent abortive DNA repair intermediates when the number of telomeric repeats is too small to completely inhibit DNA damage signaling but is sufficient to prevent the final covalent ligation step of NHEJ and induces the M1 checkpoint arrest in normal human cells. Rather than being all-or-none, telomere deprotection would thus proceed first through TAs before additional shortening leads to dicentric chromosomes. M2/crisis involves both qualitative changes (a shift from TAs to TAs plus dicentric chromosomes) and quantitative changes (an increase in the number of dysfunctional telomeres).