Free delayed recall is considered the memory measure with the greatest sensitivity for the early diagnosis of dementia. However, its specificity for dementia could be lower, as deficits other than those of pure memory might account for poor performance in this difficult and effortful task. Cued recall is supposed to allow a better distinction between poor memory due to concurrent factors and impairments related to the neurodegenerative process. The available cued recall tests suffer from a ceiling effect. This is a prospective, longitudinal study aiming to assess the utility of a new memory test based on cued recall that avoids the ceiling effect in the early diagnosis of Alzheimer's disease (AD). Twenty-five patients with mild cognitive impairment (MCI), 22 probable AD patients (NINCDS-ADRDA) at a mild stage, 22 elderly patients with subjective memory complaints (SMC) and 38 normal age-matched controls took part in the study. The patients underwent a thorough cognitive evaluation and the recommended screening procedure for the diagnosis of dementia. All patients were re-examined 12-18 months later. A newly devised delayed cued recall test using semantic cues (The RI48 Test) was compared with three established memory tests: the Ten Word-List Recall from CERAD, the "Doors" and the "Shapes" Tests from "The Doors and People Test Battery". Forty-four % of the MCI patients fulfilled criteria for probable AD at follow-up. The RI48 Test classified correctly 88% of the MCI and SMC participants and was the best predictor of the status of MCI and mild AD as well as the outcome of the MCI patients. Poor visual memory was the second best predictor of those MCI patients who evolved to AD. A cued recall test which avoids the ceiling effect is at least as good as the delayed free recall tests in the early detection of AD.