Ocular information was observed during a set of dementia tests involving participants with Alzheimer’s disease (AD), with a mild level of cognitive impairment (MCI) or in a control group. The number of participants was 26. Features of changes in pupil size and in the central position of both eyes of participants of all three types were compared. There are significant differences in some of the metrics between the types, in earlier test sessions. The possibility of classification was confirmed using the extracted features, and the contributions of some features were examined. During the Mini-Mental State Examination (MMSE) surveillance, 11-dimensional ocular information was measured in six tests. The 26 elder participants were diagnosed as Alzheimer’s disease (AD), mild cognitive impairment (MCI) and normal control (NC). The extracted features were compared between participant’s groups and tests. In order to predict participant’s group using the feature values and random forest technique, the training data set was generated using the statistic in each group. Feature selections were conducted to optimise the prediction performance. As the results, ordinal eye movement features were selected to estimate participant’s groups.