The validity of delayed-matching-to-sample (DMTS) and related tasks executed in skinner boxes as an animal model for human working memory (WM) is confounded by the occurrence of mediating behaviour during delays. True matching, a supplementary task during delays and response similarity are ways to deal with this problem. However, until now rats have not been able to learn a true matching task in Skinner boxes and introduction of a supplementary task during delays does not sufficiently prevent mediating behaviour. Response similarity, on the other hand, effectively prevents the use of mediating behaviour by reducing the discriminative value of the behaviour during delays. Furthermore, it is argued that the interpretation of drug effects is confounded by baseline performance and mediating behaviour. It is shown that high baseline levels and high amounts of mediating behaviour can induce delay dependent drug effects, suggesting a specific effect on WM. We therefore assert that examination of delay-dependency of a drug effect alone is not sufficient to claim specific effects of a drug on WM. The delayed-conditional-discrimination (DCD) task uses response similarity to effectively reduce mediating behavior and does not generate high levels of baseline performance. The DCD task is therefore preferred over other tasks for the measurement of WM in rats using Skinner boxes.