In 2 experiments, subjects engaged in 6 or 7 124-trial blocks of a continuous memory-updating paradigm with study and test trials intermixed. They studied name–location associations and were tested for the location most recently associated with a given name. On study trials, all responses were to be made on a right-hand map. On default test trials, responses were also to be made on the right-hand map, but on special test trials, designated as such during study (by showing the associations in green), responses were to be made on a left-hand map. Both default and special test trials occurred after short (2-back) and long (8-back) retention intervals. In the penultimate (Experiment 1) or last (Experiment 2) block, trials occurred under conditions in which working memory was occupied with a concurrent secondary counting backwards task. Memory for the name–location associations was better with short than with long retention intervals and was better when special (rather than default) responses were to be made, especially at the long retention interval, even with counting backwards. Thus, the intention to respond in a special way protects against forgetting associations, and this protection is not simply due to holding the information from the special trials in working memory.