We present a model of attention-based awareness. States are described by attributes, some subset of which may be forgotten at the moment of choice. A fixed attention budget is allocated to the attributes to reduce the probability of this forgetting. The model predicts (1) people focus on attributes that essentially agree with their vision of the world; (2) people focus on attributes that are unfamiliar; (3) those with a larger attention budget will make fewer mistakes. These predictions are applied to a classical preference reversal experiment. The data from the experiment include eye-tracking information that measures attention directly, allowing us to test the model's predictions. Results support the model: (1) people who choose a bet with a relatively high prize (probability) focus relatively more on that attribute; (2) overall, people focus more on probabilities than on prizes; (3) those who make reversals pay significantly less attention to the problem than those who are consistent.