Background: Despite a lot of research on the effectiveness of Pair Programming (PP), the question when it is useful or less useful remains unsettled.

Method: We analyze recordings of many industrial PP sessions with Grounded Theory Methodology and build on prior work that identified various phenomena related to within-session knowledge build-up and transfer. We validate our findings with practitioners.

Result: We identify two fundamentally different types of required knowledge and explain how different constellations of knowledge gaps in these two respects lead to different session dynamics. Gaps in project-specific systems knowledge are more hampering than gaps in general programming knowledge and are dealt with first and foremost in a PP session.

Conclusion: Partner constellations with complementary knowledge make PP a particularly effective practice. In PP sessions, differences in system understanding are more important than differences in general software development knowledge.