Tumor hypoxia is an important therapeutic target, and it can potentially be exploited by hypoxia-activated prodrugs. However, physiological hypoxia in normal tissues is a limitation. One solution would be to confine activation to severely (pathologically) hypoxic tissue, using hypoxia-activated prodrugs that provide a bystander effect through diffusion of the activated cytotoxin to adjacent regions at intermediate oxygen concentrations (associated with partial radioresistance). To evaluate this requirement, we identified five hypoxia-activated prodrugs with at least 10-fold higher potency against a cell line (A549-P540(puro)) overexpressing human cytochrome P450 reductase (P450R) relative to A549-Lo21 cells with 200-fold lower P450R activity. Bystander killing by these hypoxia-activated prodrugs was tested in anoxic multicellular layer co-cultures of these two cell lines. Cytotoxic potency against A549-Lo21 cells was unaffected by the presence of A549-P450(puro) cells for tirapazamine and RSU-1069 but increased more than 10-fold for the aziridinyldintrobenzamide CB 1954, more than 14-fold for the corresponding nitrogen mustard SN 23862, and 15-fold for its water-soluble analog SN 23816. The cytotoxic extracellular metabolites resulting from hypoxic nitroreduction of CB 1954 and SN 23862 by A549-P450(puro) cells were identified by LC/MS and bioassay methods. For SN 23862, these included the 2-amine metabolite, previously, identified as the bystander metabolite from aerobic activation by the E. coli nfsB nitroreductase, but also novel di-reduced metabolites. Cytotoxicity of SN 23862 to A549-P450(puro) cells was inhibited by lower concentrations of oxygen than for tirapazamine. The combination of selective activation under severe hypoxia with an efficient bystander effect identifies the dinitrobenzamide mustards for further development as hypoxia-activated prodrugs.