An outbreak caused by salted salmon roe contaminated with enterohemorrhagic Escherichia coli O157 occurred in Japan in 1998. Since about 0.75 to 1.5 viable cells were estimated to cause infection, we presumed that O157 might enter the viable but nonculturable (VNC) state in salted salmon roe and consequently that viable cell numbers might be underestimated. Although patient-originating O157 cells could not grow on agar plates after 72 h of incubation in 13% NaCl, they were resuscitated in yeast extract broth, and more than 90% of the cells were shown to be viable by fluorescent staining, suggesting that almost all of them could enter the VNC state in NaCl water. Roe-originating O157 was resistant to NaCl because it could grow on agar after 72 h of incubation in NaCl water, but about 20% of cells appeared to enter the VNC state. Therefore, germfree mice were infected with O157 to examine the resuscitation of cells in the VNC state and the retention of pathogenicity. O157 that originated in roe, but not patients, killed mice and was isolated from the intestine. However, these isolates had become sensitive to NaCl. O157 cells of roe origin incubated in normal media also killed mice and were isolated from the intestine, but they also became transiently NaCl sensitive. We therefore propose that bacterial cells might enter the VNC state under conditions of stress, such as those encountered in vivo or in high salt concentrations, and then revive when those conditions have eased. If so, the VNC state in food is potentially dangerous from a public health viewpoint and may have to be considered at the time of food inspection. Finally, the establishment of a simple recovery system for VNC cells should be established.