To test the validity of various models for recombination between extrachromosomal DNAs in mammalian cells, we measured recombination between a plasmid containing a herpesvirus thymidine kinase (tk) gene with an internal BamH1 linker insertion mutation (ptkB8) and a tk gene deleted at both ends (tkΔ3'Δ5'). The two DNAs shared 885 base pairs of perfect tk homology except for the interruption at the linker insertion site. Recombination events that restored the mutated insertion site to wild type were monitored by the generation of hypoxanthme-animopterine-thymidine-resistant colonies after cotransformation of Ltk⁻ cells with the two DNAs. We found that cleavage of the ptkB8 DNA at the linker insertion site was essential for gene restoration. If the tkΔ3'Δ5' DNA was ligated into mpl0 vector DNA, then recombination with the cleaved ptkB8 DNA was inefficient. In contrast, if it was excised from that vector by cleavage at flanking restriction sites, then recombination was stimulated about 150-fold. Using restriction site polymorphisms, we showed that most of the recombination events leading to restoration of the tk gene with the excised tkΔ3'Δ5' fragment involved three double-strand duplexes: two ptkB8 DNAs and one tkΔ3'Δ5' fragment. These results are much more readily explained by the single-strand annealing model of recombination than by the double-strand break repair model, and they suggest that the deficiency of the latter pathway for extrachromosomal mammalian recombination may be due, at least in part, to the obligate tripartite nature of the reaction. Finally, we measured the effect of DNA homology on the efficiency of the ptkB8-tkΔ3'Δ5' reaction. Our results showed a near-linear relationship between the efficiency of recombination and the amount of homology flanking either side of the linker insertion site. Moreover, we could detect thymidine kinase-positive transformants with as little as 10 base pairs of homology.