Several methods have been proposed to construct Statistical Shape Model (SSM) to aim image analysis using computer in field Computer Aided Diagnosis (CAD), Computer Assisted Surgery (CAS), and other medical applications by providing a prior knowledge. The major challenge for liver shape model is a high variation in geometry such as size, shape and volume between livers. In this paper, we have presented a new technique for the automatic Multi-Scale Statistical Shape Model (MS-SSM) of three-dimensional (3-D) liver from volumetric segmented images data. The procedure included both building of Spherical Harmonics shape description and the Wavelet transform. Principal Component Analysis (PCA) was applied to corresponding landmarks on the tanning shapes for performing leave-one-out test. Validation metrics, for comparing performances of the MS-SSM method against SSM, were the Hausdorff distance measure and statistical parameter of Dice Similarity Coefficient (DSC). We evaluated the performance of our proposed method against to traditional method. The results confirmed that the proposed MS-SSM technique was successful, and more accurate for liver domain. We also examined robustness of the method in liver classification. In this research classification was performed on feature vector obtained from PCA using Support Vector Machine (SVM) and k-Nearest Neighbors (k-NN) classifiers. Diagnostic accuracy was determined by leave-one-out cross-validation method and Receiver Operating Characteristic (ROC) analysis for each observer. The results showed that our proposed method to be more accurate and robust for liver discrimination.