We aimed at investigating rapid effects of plasma cortisol elevations on the episodic memory phase of encoding or retrieval, and on the strength of the memory trace. Participants were asked either to select a word containing the letter "e" (shallow encoding task) or to judge if a word referred to a living entity (deep encoding task). We intravenously administered a bolus of 20mg of cortisol either 5 min before encoding or 5 min before retrieval, in a between-subjects design. The study included only male participants tested in the late afternoon, and neutral words as stimuli. When cortisol administration occurred prior to retrieval, a main effect of group emerged. Recognition accuracy was higher for individuals who received cortisol compared to placebo. The higher discrimination accuracy for the cortisol group was significant for words encoded during deep but not shallow task. Cortisol administration before encoding did not affect subsequent retrieval performance (either for deep or shallow stimuli) despite a facilitatory trend. Because genomic mechanisms take some time to develop, such a mechanism cannot apply to our findings where the memory task was performed shortly after the enhancement of glucocorticoid levels. Therefore, glucocorticoids, through non-genomic fast effects, determine an enhancement in episodic memory if administered immediately prior to retrieval. This effect is more evident if the memory trace is laid down through deep encoding operations involving the recruitment of specific neural networks.