Operant delayed non-matching-to-position (delayed non-matching-to-position) tasks have been widely used as tests of working memory in rats, but have suffered some loss in sensitivity to differentiating selective mnemonic from non-mnemonic deficits due to floor and ceiling effects. To circumvent this problem, a novel delayed non-matching-to-position was developed in which the retention interval was adjusted on a trial-by-trial basis to hold performance accuracy at an intermediate value. The present study assessed the effects of three amnestic drugs in this delayed non-matching-to-position. Rats were administered (i.p.) NMDA receptor antagonist ((5R,10S)-(+)-5-Methyl-10, 11-dihydro-5H-dibenzo[a,d,] cyclohepten-5,10-imine (Dizocilpine or MK-801), muscarinic receptor antagonist (-)-scopolamine hydrobromide (scopolamine), or cannabinoid receptor agonist ((R)-(+)-[2, 3-Dihydro-5-methyl-3-(4-morpholinylmethyl)pyrrolo[1,2,3-de]-1, 4-benzoxazin-6-yl]-1-naphthalenylmethanone) (WIN 55, 212-2). At high doses, both MK-801 (0.12-0.25 mg/kg) and scopolamine (0.25 mg/kg) produced deficits not selective to working memory. At low doses, scopolamine (0.06-0.12 mg/kg) and MK-801 (0.06 mg/kg) produced no deficits in any mnemonic or secondary measures. WIN 55, 212-2 produced deficits at 2.0 mg/kg that were consistent with a specific impairment of working memory. Using this particular delayed non-matching-to-position revealed that consistent changes in performance accuracy at the short retention interval were evident for scopolamine and MK-801, at times in the absence of changes in response tendency, which are consistent with an interpretation that these drugs produce general deficits in reference or procedural memory. In contrast, cannabinoid-induced deficits in choice accuracy support previous reports of delay-dependent deficits. Together, these data suggest that this delayed non-matching-to-position task is able to differentiate deficit patterns of amnestic drugs, and isolate the effects of motivational side effects of drugs from working memory measurement.