Infant spatial working memory was assessed in 96 nine-month-old human infants using a modification of the delayed-response procedure. A person was the hidden stimulus, gaze was used as the response modality, the response context was a naturalistic "peek-a-boo" game, and a salient cue (i.e. curtains opening to reveal two windows) was used to evoke responses. Infants responded on most trials and performed significantly above chance in delay conditions of 10 and 20 seconds. Reinforcement affected performance, but performance was above chance even in a 0% reinforcement condition. These procedural variations boost estimates of infant spatial working memory over previous estimates and enhance the usefulness of the delayed-response procedure as a marker task for infant spatial working memory.