Gene amplification is a collection of processes whereby a DNA segment is reiterated to multiple copies per genome. It is important in carcinogenesis and resistance to chemotherapeutic agents, and can underlie adaptive evolution via increased expression of an amplified gene, evolution of new gene functions, and genome evolution. Though first described in the model organism Escherichia coli in the early 1960s, only scant information on the mechanism(s) of amplification in this system has been obtained, and many models for mechanism(s) were possible. More recently, some gene amplifications in E. coli were shown to be stress-inducible and to confer a selective advantage to cells under stress (adaptive amplifications), potentially accelerating evolution specifically when cells are poorly adapted to their environment. We focus on stress-induced amplification in E. coli and report several findings that indicate a novel molecular mechanism, and we suggest that most amplifications might be stress-induced, not spontaneous. First, as often hypothesized, but not shown previously, certain proteins used for DNA double-strand-break repair and homologous recombination are required for amplification. Second, in contrast with previous models in which homologous recombination between repeated sequences caused duplications that lead to amplification, the amplified DNAs are present in situ as tandem, direct repeats of 7-32 kilobases bordered by only 4 to 15 base pairs of G-rich homology, indicating an initial non-homologous recombination event. Sequences at the rearrangement junctions suggest nonhomologous recombination mechanisms that occur via template switching during DNA replication, but unlike previously described template switching events, these must occur over long distances. Third, we provide evidence that 3'-single-strand DNA ends are intermediates in the process, supporting a template-switching mechanism. Fourth, we provide evidence that lagging-strand templates are involved. Finally, we propose a novel, long-distance template-switching model for the mechanism of adaptive amplification that suggests how stress induces the amplifications. We outline its possible applicability to amplification in humans and other organisms and circumstances.