A conflict model is described in which clinically effective antianxiety agents exhibit pronounced anticonflict activity. Male squirrel monkeys were trained to depress a bar for 5 sec to obtain food reinforcement. The 6 hr test session was comprised of an initial 3 hr period in which each 5 sec response was punished and then a 3 hr unpunished period. Trained monkeys would rarely be shocked and would make most of their responses during the non-punished period. Both benzodiazepine (chlordiazepoxide and diazepam) and non-benzodiazepine (meprobamate and phenobarbital) anxiolytics produced pronounced and unequivocal increases in punished responding. Other psychoactive agents (damphetamine, chlorpromazine, ethanol, morphine, amitriptyline and imipramine) did not produce an increase in punished responding. Sensitivity (i.e., large magnitude effects), selectivity, stable baseline performance and fully automated features make this test useful in identifying potential anxiolytic agents in primates.