If someone falls off of a 14th story roof, very predictably death will result. The conditions under which readers appear to infer such predictable outcomes were examined with three different retrieval paradigms: immediate recognition test, cued recall, and priming in word recognition. On immediate test, responses to a word representing the implicit outcome (e.g., dead) were slow, but on delayed test these responses were slow or inaccurate only when primed by an explicitly stated word. However, the word expressing the predictable outcome did function as an effective recall cue. Results suggest that readers encode these inferences into memory only minimally, but that they can make use of a cue word that represents the inference (e.g., dead) both at the time of an immediate test and in delayed cued recall.