Most education campaigns for melanoma (MM) detection in the general population have used the “ABCD” algorithm, although recognition of objects in the real life is based on a holistic image recognition rather than on analytic criteria. The objective was to compare analytic (ABCD) and cognitive (photographs) strategies for teaching self‐recognition of MM. A prospective 4‐arm stratified randomized trial in 255 individuals compared 3 realistic educative interventions by leaflets: 1) ABCD algorithm (“ABCD”), 2) a set of photographs chosen to stimulate recognition of MM among benign pigmented lesions (“Cog”), 3) photographs + explanations (“Cog‐Ex” arm) and 4) no intervention (“NI”). A 40‐slides test was performed before intervention (T0), 1 week after (T1) and after induction of anxiety (T2). In the “ABCD” arm, sensitivity slightly improved (80 to 83.8%, p = 0.04), but specificity dropped from 65.1 to 56.3% (p < 0.001), with no benefit in accuracy as compared to “NI”. In “Cog”arm, there was no change in sensitivity, but a strong increase in specificity (65.9 to 81.1%, p < 0.001) and accuracy (42.1 to 53.1%, p < 0.001). “Cog‐ex” resulted in similar although lower benefit. Under stress (T2), there was a dramatic loss of specificity and accuracy in “ABCD”arm (65.1 to 44.1%, p < 0.001 and 40.8% to 35.8%, p ≤ 0.001) without higher gain in sensitivity, while sensitivity and accuracy increased (p < 0.005) after “Cog” leaflet, without decreasing specificity. Finally, the “ABCD” message alone does not seem efficacious and is even worse in the context of anxiety, whereas a quick look at a few photographs is sufficient to improve the ability of the laymen to recognize a MM just by optimizing their spontaneous image recognition capacities. Education by photographs is a realistic strategy which should replace or complete “ABCD” message in the campaigns for self‐detection of MM. © 2005 Wiley‐Liss, Inc.