Ionotropic γ-aminobutyric acid (GABA A) receptors control the relay of nociceptive signals at several levels of the neuraxis. Experiments with systemically applied benzodiazepines, which enhance the action of GABA at these receptors, have suggested both anti- and pronociceptive effects. The interpretation of such experiments has been notoriously difficult because of confounding sedation. Here, we have used genetically engineered mice, which carry specific benzodiazepine-insensitive GABA A receptor subunits, to test whether diazepam, a frequently used classical benzodiazepine, exerts antihyperalgesia after systemic administration in the formalin test, a model of tonic nociception. In wild-type mice, systemic diazepam (3–30 mg/kg, p.o.) dose-dependently reduced the number of formalin-induced flinches during both phases of the test by about 40–70%. This antinociception was reversed by the benzodiazepine site antagonist flumazenil (10 mg/kg, i.p.), but fully retained in GABA A receptor α1 point-mutated mice, which were resistant against the sedative action of diazepam. Experiments carried out in mice with two diazepam-insensitive subunits (α1/α2, α1/α3 and α1/α5 double point-mutated mice) allowed addressing the contribution of α2, α3 and α5 subunits to systemic diazepam-induced antihyperalgesia in the absence of sedation. The relative contributions of these subunits were α2 ≈ α3 > α5, and thus very similar to those found for intrathecal diazepam (0.09 mg/kg). Accordingly, SL-651498 (10 mg/kg, p.o.), an “anxioselective” benzodiazepine site agonist with preferential activity at α2/α3 subunits, significantly reduced formalin-induced flinching in wild-type mice. We conclude that systemic diazepam exerts a genuine antihyperalgesic effect, which depends on spinal GABA A receptors containing α2 and/or α3 subunits.