In order to decrease the cost, it is recommended to detect defects in the phase they occur. This paper presents a metrics-based approach to detect defects in OO designs. We first identify a list of OO design defects which have a significant impact on the design quality based on the violations of many design guidelines and best practices expressed by experts. Then, we define the metrics for automatically detecting each of these defects. Several number of well-known metrics are used. For some defects, however, there are no suitable design metrics. Therefore, we define new design metrics to detect these defects. Moreover, thresholds to judge whether a metrics value indicates a critical situation, or not are defined for each of the metrics. By defining metrics on each design defect, many design rules and heuristics, and flaws which are described qualitatively can be evaluated quantitatively. On the other hand, we will show that intended application of metrics becomes clear.