The promise of subtype-selective GABA(A) receptor drugs with anxiolytic properties but with a much reduced side-effect burden (compared to benzodiazepines) is an attainable goal. However, its achievement necessitates the availability of in vivo preclinical assays capable of demonstrating differences as well as similarities between subtype-selective agents and non-selective benzodiazepines. In this study, we have compared three mouse strains (NMRI, C57BL/6J and DBA/2) in four models of anxiety-like behaviour (plus-maze, zero-maze, light-dark, and Vogel conflict). Furthermore, in each model, we have contrasted in detail the behavioural responses of each strain to the non-selective benzodiazepine chlordiazepoxide (CDP; 5-20 mg/kg), and the subtype-selective agents L-838,417 (GABA(A)-alpha(2/3/5); 3-30 mg/kg) and zolpidem (GABA(A)-alpha1; 0.3-3.0 mg/kg). The data show a complex mouse strainxmodelxpharmacological agent interaction. Most importantly, not all mouse strainxmodel test systems showed a positive response to CDP or predicted the response to L-838,417. This dissociation between CDP and L-838,417 opens up opportunities for preclinical test systems that differentiate subtype-selective and non-selective GABA(A) receptor agents, an attribute that might well be important in providing the necessary confidence for further drug development. Present findings suggest the need for a much greater focus on defining test systems appropriate for screening novel chemical entities, rather than self-selection of models or genotypes based on responses to known pharmacological agents. For example, if current data with L-838,417 are confirmed with compounds showing similar selectivity profiles, such agents may in future be best identified and characterised using test systems comprising NMRI mice in the zero-maze and/or C57 mice in the Vogel conflict and/or light-dark tests.