Replication fork maintenance pathways preserve chromosomes, but their faulty application at nonallelic repeats could generate rearrangements causing cancer, genomic disorders and speciation(1-3). Potential causal mechanisms are homologous recombination and error-free postreplication repair (EF-PRR). Homologous recombination repairs damage-induced DNA double-strand breaks (DSBs) and single-ended DSBs within replication. To facilitate homologous recombination, the recombinase RAD51 and mediator BRCA2 forma filament on the 3' DNA strand at a break to enable annealing to the complementary sister chromatid(4) while the RecQ helicase, BLM (Bloom syndrome mutated) suppresses crossing over to prevent recombination(5). Homologous recombination also stabilizes(6,7) and restarts(8,9) replication forks without a DSB10,11. EF-PRR bypasses DNA incongruities that impede replication by ubiquitinating PCNA (proliferating cell nuclear antigen) using the RAD6-RAD18 and UBC13-MMS2-RAD5 ubiquitin ligase complexes(12). Some components are common to both homologous recombination and EF-PRR such as RAD51 and RAD18(13,14). Here we delineate two pathways that spontaneously fuse inverted repeats to generate unstable chromosomal rearrangements in wild-type mouse embryonic stem (ES) cells. Gamma-radiation induced a BLM-regulated pathway that selectively fused identical, but not mismatched, repeats. By contrast, ultraviolet light induced a RAD18-dependent pathway that efficiently fused mismatched repeats. Furthermore, TREX2 (a 3'-> 5' exonuclease) suppressed identical repeat fusion but enhanced mismatched repeat fusion, clearly separating these pathways. TREX2 associated with UBC13 and enhanced PCNA ubiquitination in response to ultraviolet light, consistent with it being a novel member of EF-PRR. RAD18 and TREX2 also suppressed replication fork stalling in response to nucleotide depletion. Interestingly, replication fork stalling induced fusion for identical and mismatched repeats, implicating faulty replication as a causal mechanism for both pathways.