The fate of a bacterium and two yeast species genetically engineered by insertion of a nucleotide sequence encoding for aprotinin was studied in three different soils. Corynebacterium glutamicum carried the recombinant gene on plasmid pUN1, Saccharomyces cerevisiae carried the gene on plasmid p707, and in Pichia angusta (formerly Hansenula polymoropha) LR9-Apr8, the gene was chromosomally inserted with eight tandem repeats. Corynebacterium glutamicum persisted longer than both yeast strains. In a sandy loam of pH 5.9, recovery rates of cultured cells were lower than in a clay silt or a silty sand, with pH values of 7.1 and 6.7, respectively. Generally, persistence at 10 °C was higher than at 20 °C. An adaptation of the genetically engineered strains resulting in higher soil persistence was not observed for any of the three species tested. Competition experiments between nonengineered and genetically engineered strains in presterilized soils revealed a reduced fitness of the engineered strains. However, a more competitive C. glutamicum pUN1 evolved after reinoculation of cells, preselected by a preceding competition experiment.Key words: ecological risk assessment, genetic engineering, nondeliberate release, soil inoculation, aprotinin.