Computer Aided Diagnosis (CAD) is useful to detect Alzheimer's disease (AD) at an early stage. For detecting the presence of Alzheimer's disease Magnetic Resonance Imaging is very useful. In this paper, a novel method for distinguishing between Mild cognitive impairment (MCI), Normal control (NC) and Alzheimer disease (AD) subjects is proposed based on Multivariate techniques such as Partial Least Square (PLS) and Principal Component Analysis (PCA). Median filtering is preferred due to its excellent noise removal characteristics. K-means clustering is preferred to segment the image into different anatomical structures and to yield better segmentation. Feature extraction is done using Haralick texture parameters. Support Vector Machine (SVM) classifier is used to validate the efficiency and performance of the proposed method.