Four studies examined how people combine social concepts that have conflicting implications (e.g., Harvard-educated and carpenter). Several kinds of evidence indicated that such combinations are guided by causal reasoning that draws upon both causal relations contained within the constituent concepts and on broader world knowledge. Open-ended descriptions of members of combinations contained explicit causal descriptors, as. well as emergent attributes not used to describe members of constituents. Ratings of the likelihood that combination members possessed various attributes were not fully predicted by comparable ratings of constituents. Causal reasoning appeared to be most pervasive for combinations viewed as more surprising, suggesting that surprise may have triggered the generation of causal accounts.