Background: At 3 years after diagnosis, the risk of Alzheimer disease (AD) for patients with mild cognitive impairment (MCI) is estimated to be 18% to 30%. To improve treatment of patients at high dementia risk there is a need for a better prediction of the risk for transition from MCI to AD. Olfactory deficits are a hypothetical predictor of conversion form MCI to AD. Furthermore, several studies point at volumetric reduction of medial temporal lobe structures as predictors of conversion form MCI to AD. The primary aim of this study was to evaluate whether investigations of odor deficits in MCI combined with neuropsychological tests and MRI examinations can improve prediction of the development of dementia. Methods: Changes in olfactory functions, cognitive functions, and volume of medial temporal lobe structures (hippocampus, parahippocampal gyrus, and amygdala) were evaluated in a 24-month follow-up study in 49 MCI patients and 33 controls. Results: In the MCI group, a prediction of strong cognitive functions deterioration based on poor performance in Olfactory Identification tests shows sensitivity of 57% and specificity of 88%. The test based on cognitive functions only shows a sensitivity of 44%, and 89%, respectively. Combined tests having a criteria of poor olfactory identification performance AND poor results of neuropsychological tests showed a sensitivity of 100% and specificity of 84%. Furthermore, correlation was found between the results of Olfactory Identification tests at baseline and deterioration of cognitive functions at follow up. Odor identification threshold did not appear to be a dementia predictor. A correlation of progress of cognitive function deterioration, odor identification deterioration, and decrease of volume of the hippocampus was also observed. Conclusions: Prediction of MCI to dementia conversion can be improved by supplementing the neuropsychological tests with odor identification tests. A follow up study of hippocampus volume reduction, OI performance and cognitive functions deterioration will further increase prediction accuracy.