Episodic memory impairment is a hallmark for early diagnosis of Alzheimer's disease. Most actual tests used to diagnose Alzheimer's disease do not assess the spatiotemporal properties of episodic memory and lead to false-positive or -negative diagnosis. We used a newly developed, nonverbal navigation test for Human, based on the objective experimental testing of a spatiotemporal experience, to differentially Alzheimer's disease at the mild stage (N = 16 patients) from frontotemporal lobar degeneration (N = 11 patients) and normal aging (N = 24 subjects). Comparing navigation parameters and standard neuropsychological tests, temporal order memory appeared to have the highest predictive power for mild Alzheimer's disease diagnosis versus frontotemporal lobar degeneration and normal aging. This test was also nonredundant with classical neuropsychological tests. As a conclusion, our results suggest that temporal order memory tested in a spatial navigation task may provide a selective behavioral marker of Alzheimer's disease.