Objective and early detection of Alzheimer's disease (AD) is a demanding problem requiring consideration of manymodal observations. Potentially, many features could be used to discern between people without AD and those at different stages of the disease. Such features include results from cognitive and memory tests, imaging (MRI, PET) results, cerebral spine fluid data, blood markers etc. However, in order to define an efficient and limited set of features that can be employed in classifiers requires mining of data from many patient cases. In this study we used two databases, ADNI and Kuopio LMCI, to investigate the relative importance of features and their combinations. Optimal feature combinations are to be used in a Clinical Decision Support System that is to be used in clinical AD diagnosis practice.