The effects of flumazenil (Ro 15-1788) and a new beta-carboline, ambocarb (AMB), on learning were investigated using the multichoice maze. The drugs, administered either alone or simultaneously, were injected once a day before training for eight days. AMB, administered alone, improved the performance and decreased the working errors, whilst flumazenil had no effect on performance during its sole administration but weakly prevented the learning-improving effect of AMB. More significantly, flumazenil antagonized the motor activity depressed by AMB. In the study ex vivo, flumazenil decreased and AMB increased the apparent affinity of [3H]flunitrazepam to the central benzodiazepine receptors. Flumazenil reversed the action of AMB on the central benzodiazepine receptors, but failed to reduce significantly the modulative effects of AMB on [3H]muscimol and [35S]t-butylbicyclophosphorothionate ([35S]TBPS) binding. These data indicate that flumazenil, due to its action on the central benzodiazepine receptors, more effectively reverses the inhibition of motor activity than the performance-improving effect of AMB.