At our institution, the prevalence of clinical isolates of Clostridium difficile with resistance to metronidazole is 6.3%. We observed that initial metronidazole MICs of 16 to 64 mg/liter against toxigenic, primary fresh C. difficile isolates, as determined by agar dilution, decreased to 0.125 mg/liter after the isolates were thawed. In this study, we examined the possibility of heterogeneous or inducible resistance. Totals of 14 metronidazole-resistant and 10 metronidazole-susceptible clinical isolates of toxigenic C. difficile were studied. The isolates were investigated for the presence of nim genes by PCR. After the isolates were thawed, susceptibility testing was done by agar dilution, by disc diffusion using a 5-mug metronidazole disc, and by the Etest method. An experiment for determining the effect of prolonged exposure to metronidazole was applied to all resistant isolates and to susceptible control strains. None of the isolates presented the nim genes. All initially metronidazole-resistant C. difficile isolates became susceptible after thawing; however, they presented slow-growing subpopulations within the inhibition zones of both the disk and the Etest strip. All metronidazole-susceptible isolates remained homogeneously susceptible by both methods. After prolonged exposure in vitro to metronidazole, no zone of inhibition was found around the 5-microg disk in any of the metronidazole-resistant isolates, and the MICs as determined by the Etest method ranged from 0.125 to >256 mg/liter, with colonies growing inside the inhibition zone. Our results indicate that (i) resistance to metronidazole was not due to the presence of nim genes, (ii) resistance to metronidazole in toxigenic C. difficile isolates is heterogeneous, and (iii) prolonged exposure to metronidazole can select for in vitro resistance. We recommend routine performance of the disk diffusion method (5-microg metronidazole disk) with primary fresh C. difficile isolates in order to ensure that metronidazole-heteroresistant populations do not go undetected.