Object-oriented software metrics provide a numerical characterization of software quality. They have also been used in the assessment and identification of technical debt. However, metrics generally need to be used with thresholds as reference points that help to interpret their values properly and objectively. The problem is that, while there are many proposed metrics, there are relatively few studies on thresholds and threshold calculation methods; hence, the effective application of metrics in practice has been limited. Moreover, although it has been acknowledged that thresholds should not be absolute, but rather relative to certain contextual factors, the context is still not considered in most threshold studies. In this paper, the relationship between system size (as a contextual factor) and metric thresholds is investigated. The objective is to build predictive models that estimate thresholds based solely on system size, and to assess the feasibility of this approach as a threshold estimation method. An empirical study is conducted for this purpose using 36 defect-prediction datasets and six metrics. The results show that the proposed threshold estimation method is feasible, and it can achieve an accuracy remarkably comparable to more complex threshold models.