We studied the integrity of working memory operations in 38 HIV-seropositive and 20 seronegative drug users, using a modified version of the Tower of London task. This new task, the Tower of London–Working Memory version (TOL–WM), includes a delayed-response component in addition to the planning required for successful performance of the standard TOL. Symptomatic HIV-seropositive participants solved significantly fewer TOL–WM problems compared to matched seronegative controls. However, seropositive and seronegative subjects showed similar overall levels of planning efficiency, suggesting that the TOL–WM deficit may be associated primarily with failure to encode or maintain an adequate online memory representation. The results of this study confirm our previous report of a possible working memory deficit in HIV-1 infection and suggest that measures of working memory have particular utility in the evaluation of HIV-related cognitive deficits. (JINS, 1997, 3, 451–456.)