We compared the stability of antibiotic-resistance markers in strains derived from Bacillus cereus UW85 in culture media and in the soybean rhizosphere in a growth chamber and in the field. We studied two independent, spontaneous mutants resistant to neomycin, three independent, spontaneous mutants resistant to streptomycin, and strains carrying plasmid pBC16, which encodes tetracycline resistance. Antibiotic-resistance markers were maintained in populations of all UW85 derivatives in culture and in the rhizosphere of soybeans grown in soil in a growth chamber. In two field experiments, antibiotic resistance was substantially lost in rhizosphere populations of B. cereus as early as 14 or as late as 116 days after planting. To distinguish between death of the inoculated strain and loss of its marker, we tested populations of B. cereus for other phenotypes (orange pigmentation, plasmid-borne resistance to tetracycline, and biocontrol activity) that are typical of UW85-derivatives used as inoculum, but atypical of the indigenous populations of B. cereus, and these phenotypes were maintained in populations from which the marker was lost. In general, neomycin-resistance markers were maintained at a higher frequency than streptomycin-resistance markers, and maintenance of antibiotic-resistance markers varied with position on the root and with the year of the experiment. In a semi-defined medium, the UW85 derivatives grew at the same rate as the wild type at 28 degrees C, but most grew more slowly than the wild type at 16 degrees C, demonstrating that antibiotic resistance can affect fitness under some conditions. The results suggest that the stability of antibiotic-resistance markers should be assessed in the ecosystems in which they will be studied.