Uncertainty and learning are key components to many environmental externalities. Often the true costs of pollution, be they from greenhouse gases or deforestation are unknown at the time the pollution is created, and policy makers need to decide on mitigation before they know the full extent of the damage. In this paper, we ask two related questions: (1) What is the effect of the potential for learning on the timing and amount of investment and (2) In which environmental policy situations will the potential for learning lead to an increase in initial mitigation? By explicitly modeling the structure of information, and treating learning as a continuous variable, we derive a simple condition that dictates when the prospect of learning will increase initial mitigation, namely, when the curvature elasticity of the marginal cost of mitigation is at least twice as large as the curvature elasticity of marginal benefit. The lower the amount of anticipated learning, the higher the ratio of curvature elasticity of marginal cost to benefit required for this 'precautionary' result. Facing a discount rate exacerbates the required ratio of curvature elasticities, while the introduction of a small stock externality makes it more likely that learning will increase the initial optimal level of mitigation.