The usage of software metrics for various purposes has become a hot research topic in academia and industry (e.g. detecting design patterns and bad smells, studying change-proneness, quality and maintainability, predicting faults). Most of these topics have one thing in common: they are all using some kind of metrics-based models to achieve their goal. Unfortunately, only few researchers have tested these models on unknown software systems so far.

This paper tackles the question, which metrics are suitable for preparing portable models (which can be efficiently applied to unknown software systems). We have assessed several metrics on four large software systems and we found that the well-known RFC and WMC metrics differentiate the analyzed systems fairly well. Consequently, these metrics cannot be used to build portable models, while the CBO, LCOM and LOC metrics behave similarly on all systems, so they seem to be suitable for this purpose.