Two different approaches for introducing pathogenic mutations into the -amyloid precursor protein gene in mouse embryonic stem cells were compared. Both approaches require two sequential modifications of the targeting locus by homologous recombinations. One approach was a "targeting-in-out" procedure that is based on a double-replacement strategy, and the other was a "hit-and-run" procedure that makes use of an unstable genomic duplication after vector integration. Both approaches showed similar targeting frequencies for the first step. In the targeting-in-out procedure, targeted-in embryonic stem cell clones with the desired mutation and an intron-located selection cassette were obtained at a high frequency after the first step. Targeting out, however, resulted not only in the expected loss of the intron-located selection cassette but also in unavoidable reversion to wild type. In contrast, pure mutants, i.e., those without additional genomic changes, were generated by the hit-and-run procedure. Although targeted-in embryonic stem cells might be used to generate animals with modified -amyloid precursor protein, the hit-and-run procedure appears to be the superior way to target gene modifications in vivo, leading to pure, correct mutants. For further improvements, optimization of the homologous recombination efficiency could be envisaged.