Abstract: Almost every general model of constructive memory has a lot to say about how generalizations are abstracted, information is synthesized and patterns are discovered. In other words, these accounts are about combining, compressing and drawing connections. However, many memory operations are better thought of as dividing and isolating. The paradigmatic example is a memory anomaly. For instance, you might remember a strange interaction with a friend where their behaviour seemed out of character, only to later discover that they had witnessed a traumatic event on that day. What is the purpose of keeping these odd memories around and how do we select for them? How does this process relate to the more familiar one of systematizing and combining? This paper compares information-theoretic and informal measures of what makes an event anomalous and proposes a preliminary theory of the kinds of weird events that should stick in our memory.