An object-oriented approach has become a commonly-used method in software-related activities. Many design metrics for object-oriented systems have been proposed and also employed for predicting and managing the quality of processes and products. To enhance the practical utility of object-oriented metrics in software industry, various researchers have tried to find relations between these metrics and fault proneness, but very few focus on relating them with the number-offaults in different levels as per their severity rating. In this study, empirical validation is carried out on object-oriented design metrics (i.e. Chidamber and Kemerer CK-metrics suite and source lines of codes) for predicting number-of-faults in different severity levels. Different statistical methods are used to analyze the data, including correlation.