Working memory was modeled in rats using a delayed response (DR) task with spatial location as the discriminative cue. Rats received food for pressing 1 of 2 retractable levers in the choice phase of a trial if that lever had been presented in the prior sample phase of that trial. When delays of 0 - 20 sec were imposed between sample and choice, choice accuracy declined with increasing delay. With short intertrial intervals (ITIs), choice accuracy decreased more at long delays than at short delays, showing that interference from previous trials impaired memory but not discrimination. Trimethyltin (TMT), 7 mg/kg iv, reduced the choice accuracy of one rat to chance levels at all delays; two other rats were affected transiently. TMT did not affect responses to the retracted levers during delays. TMT treatment also elevated levels of glial fibrillary acid protein (GFAP) in the CNS, measured 4 weeks after treatment. GFAP levels also correlated significantly with TMT-induced changes in the slopes and intercepts of the retention gradients for individual animals at both times after treatment. The delay and ITI effects suggest that the DR task adequately assessed memory. The lack of effect of delay presses on choice accuracy suggests that these overt mediating responses did not differentially affect choice responding. The preferential disruption of choice accuracy at long delays by TMT confirms previous reports that working memory is mediated by the limbic system. The lack of effect of TMT on delay presses suggests that its neurotoxic effect on memory did not involve these mediating responses.