The use of benzodiazepines (BDZs) in the long-term treatment of epilepsy is greatly restricted by their capacity to induce tolerance and dependence. Thus, the development of new BDZ-related therapeutic agents should be directed by strategies that minimize tolerance- and dependence-inducing properties. Experimental procedures used to determine the success of such strategies often rely on a single assay procedure (e.g., one seizure model), which might lead to false predictions. Furthermore, the different types of tolerance, i.e., "pharmacological" (metabolic or functional) and "behavioral" ("learned" or "contingent"), are often not dealt with in such studies. This prompted us to compare the chronic anticonvulsant efficacy and withdrawal characteristics of diazepam and two novel BDZ receptor ligands, i.e., the partial agonist bretazenil and the subtype-selective agonist abecarnil, in different seizure models in mice. Myoclonic, clonic and tonic seizures were induced by i.v. infusion of pentylenetetrazol and by transcorneal or transauricular application of electrical stimuli. Prolonged administration of diazepam (5 mg/kg twice daily for 6 days) resulted in marked anticonvulsant effects on myoclonic, clonic and tonic seizure thresholds at the onset of treatment, but pronounced tolerance developed rapidly during subsequent treatment. The time course and extent of tolerance was similar with most seizure models. Tolerance characteristics were not affected by study design, i.e., use of separate or the same animals for each seizure induction, indicating that learned or contingent tolerance was not significantly involved under these experimental conditions. After termination of treatment with diazepam, significant seizure threshold decreases were determined, indicating withdrawal hyperexcitability in response to physical dependence. During prolonged administration of abecarnil (10 mg/kg twice daily for 6 days), some anticonvulsant tolerance was seen with electroshock seizures, but not with pentylenetetrazol seizures; no withdrawal hyperexcitability was determined upon termination of treatment. Bretazenil (10 mg/kg twice daily for 6 days) produced no tolerance in any of the seizure models, but a significant decrease in electroshock seizure threshold was seen in the withdrawal period. The data indicate that tolerance and withdrawal characteristics of BDZ receptor partial and subtype-selective agonists in mice depend on the experimental model used, whereas the influence of the experimental protocol is less critical in the case of a full BDZ receptor agonist such as diazepam.