Pretend play is often hypothesized in a global sense to be an effective context for young children's learning, but there is much still to learn about whether all types of information can be learned equally and whether all types of pretend play are equally beneficial. The present study tests whether preschoolers can learn a simple, novel causal rule from a pretend demonstration. American children watched either one (Experiment 1) or two (Experiment 2) experimenters pretend that a wooden box was a "blicket machine" and that certain blocks were "blickets." In Experiment 1, performance was compared to a control group who saw an experimenter use the same materials in a realistic context. Results show that participants in the control condition outperformed children in the pretend condition in learning the novel causal rule. Data from both experiments combined revealed that, in pretense, participants learned the novel causal rule when the rule involved block color but not when it involved block shape. These results add to previous research on learning from pretense which indicates that learning novel information from pretense is selective and not always as effective as learning from real contexts.