Non-selective benzodiazepine (BZ) binding-site full agonists, exemplified by diazepam, act by enhancing the inhibitory effects of GABA at GABA(A) receptors containing either an alpha1, -2, -3 or -5 subunit. However, despite their proven clinical anxiolytic efficacy, such compounds possess a relatively narrow window between doses that produce anxiolysis and those that cause sedation, and are also associated with physical dependence and a potential for abuse. In the late 1980s and early 1990s a number of non-selective partial agonists, exemplified by bretazenil, pazinaclone and abecarnil, were described. Their reduced intrinsic efficacy relative to full agonists such as diazepam resulted in an improved preclinical pharmacological profile in that there was a large window between anxiolytic and sedative doses and their dependence and abuse liabilities were much lower. Unfortunately, these compounds failed, for a variety of reasons, to translate into clinical benefit, and as the public perception of BZs deteriorated interest in the area waned. However, the advent of molecular genetic and pharmacological approaches has begun to delineate which GABA(A) receptor subtypes are associated with the various pharmacological effects of the non-selective BZs. More specifically, the alpha2- and/or alpha3-containing GABA(A) receptors play a role in anxiety whereas the alpha1 subtype is involved in sedation, raising the possibility of a compound that selectively modulates alpha2- and/or alpha3-containing receptors but does not affect alpha1-containing receptors would be a non-sedating anxiolytic. In order to achieve selectivity for the alpha2/alpha3 subtypes relative to alpha1, two approaches may be used; selective affinity or selective efficacy. Selective affinity relies on a compound binding with higher affinity to the alpha2/alpha3 compared with alpha1 subtypes, but to date no such compounds have been described. On the other hand, subtype-selective efficacy relies on a compound binding to all subtypes but having different efficacies at various subtypes (relative selective efficacy, for example SL654198 or pagoclone) or having efficacy at some subtypes but none at others (absolute selective efficacy; for example, L-838417). The status of these and other BZ site compounds with claimed, but often not explicitly stated, GABA(A) subtype selectivity (such as ELB-139 and ocinaplon) will be reviewed in relation to their development as non-sedating anxiolytics for the treatment of generalised anxiety disorder.