BACKGROUND The visual association test (VAT) is a brief learning task based on imagery mnemonics. The test materials consist of six line drawings of pairs of interacting objects or animals-for example, an ape holding an umbrella. The person is asked to name each object and, later, is presented with one object from the pair and asked to name the other. OBJECTIVE To verify that the task induces robust incidental or effortless learning (study 1), and to study the efficiency of the test as a discriminator between early dementia of the Alzheimer type (DAT) and non-demented people (study 2) and non-DAT types of dementia (study 3). METHODS Study 1: two groups of elderly volunteers were administered the VAT. The stimuli were presented in the interactive fashion to group A-for example, a monkey carrying an umbrella (n=83)-and side by side to group B-for example, separate pictures of a monkey alone and an umbrella alone (n=79). Group B received learning instructions, but group A did not. Study 2: three groups of subjects were selected from a population based follow up study: incident DAT cases (n=24), cognitively declining subjects not diagnosed with dementia (n=21), and stable non-demented subjects (n=204). Test performance of the non-demented group at baseline was compared with that of patients with DAT at the time of their diagnosis, of patients with DAT a year before their diagnosis, and of non-demented declining subjects at baseline. Study 3: subjects were patients referred for neuropsychological assessment because of suspected dementia. They were diagnosed by consensus criteria of various dementia syndromes. RESULTS Study 1: recall was more than twice as high in group A as in group B. Thus interactive presentation, even in the absence of learning instructions, enhances learning. Study 2: at a level of 97.5% specificity, the VAT had a sensitivity of 87.5% for DAT cases at the time of diagnosis and 66.7% one year before diagnosis. The cognitively declining group scored significantly lower on the VAT at baseline than the non-demented group. The VAT discriminated more effectively than both the MMSE and the six item picture learning task from the CAMCOG. Study 3: VAT scores were significantly lower in patients with DAT (n=48) than in patients with vascular dementia (n=37), frontotemporal dementia (n=9), or subcortical dementia (n=15), but not lower than in patients with Lewy body dementia (n=7). Mean mini mental state examination scores of these groups were not significantly different. The VAT discriminated patients with DAT from patients with other types of dementia more effectively than a prose recall test. Sensitivity was 79% and specificity 69%. CONCLUSIONS The VAT detects with high specificity a sizeable proportion of patients with DAT a year before the diagnosis, and a low VAT score is relatively uncommon in patients with non-DAT dementia.