The effects of peripherally injected arginine vasopressin (AVP: 0-25 micrograms/kg), its desglycinamide analogue (DGAVP: 0-25 micrograms/kg), which is practically devoid of pressor activity, and d-amphetamine (AMP: 0-1.25 mg/kg) were studied using a delayed (0-32 s) matching to position task (Dunnett 1985). A limited hold for responding (20 s) was in operation. This task enables an accurate assessment of forgetting in rats. AVP reliably improved per cent correct performance, and this effect was substantiated by accuracy indices derived from signal detection theory (TSD). DGAVP, however, was inactive, suggesting that the parent peptide's pressor properties were responsible for its beneficial effects. AMP disrupted performance in a dose-related manner, and was the only substance to alter a TSD bias index (responsivity index, RI), indicating a degree of response repetition at the highest dose. These results are consistent with some earlier reports, and suggest that AVP may enhance memory by peripheral action, while AMP disrupts performance. Closer inspection of the data, however, suggested that the peptide reduced general responsiveness. A new index to measure bias (Sahgal 1987) suggested that AVP-treated subjects restricted their sample, and choice responses to one side of the operant chamber, thereby achieving a spuriously high detection rate with few errors of commission (incorrect responses). It is concluded that AVP does not, after all, improve performance: on the contrary it has detrimental effects, and produce errors of omission (failure to respond).