Although borrowing and contact are recognised as important factors in language histories, there is no clear and agreed way of dealing with their effects on methods like traditional lexicostatistics. We argue that one promising approach involves subdividing standard meaning-lists into more and less conservative sublists. Differences between trees or networks generated from these sublists may then indicate borrowing. This approach has been tested on Indo-European, but here we apply it to languages of the Andes in an attempt to answer the vexed Quechumara question: are Quechua and Aymara genetically related, or linked only by contact? Our evidence suggests that contact is the more likely explanation for parallels between Quechua and Aymara.