The generation of predictive inferences may be difficult when a story leads to multiple possible consequences. The present study examined whether inferences are generated when the story implies two mutually exclusive consequences are nearly equally likely to occur. Experiment 1 used a word-naming task and showed that neither inference was activated when probed at 500 ms after the story, suggesting the alternative inference interfered with activation of the targeted inference. Experiment 2 used a contradictory reading paradigm and showed that reading time was slower when a sentence contradicted both inferences but not when it contradicted only one inference. These results imply that both inferences were generated but at a minimal level of activation so that they were not detected in a word-naming task. The results are discussed in the light of comprehension theories that could account for the representation of minimally encoded inferences.