Traditionally, climate modelers try to make their computer simulations reflect physical reality in as much detail as possible. That approach is wonderful for helping scientists probe the mechanics of Earth's climate. What it's not so good at is predicting the future—the very function that policymakers need most. Some researchers say the only answer is a major overhaul. Climate models, they say, need to become more like models used in weather forecasting and economics, soft-pedaling physics and stressing probability. Such a "stochastic" approach would mark a sea change in climatology; naturally, not everyone in the field is sold on the idea.