In providing precautionary incentives and allocating environmental risk, liability for environmental damages faces two difficulties: randomness of damages incurred and nonobservability of precautionary measures. Hence, a major obstacle in enforcing legal claims is the difficulty of proving causation. In order to alleviate the proof of causation, the German Umwelthaftungsgesetz introduces a refutable causality presumption. When imperfect information is taken into account, this conditional reversal of the burden of proof cannot ensure optimal precaution. Using a principalagent model, optimality conditions for a modified strict liability rule is derived. A specific rule for the optimal allocation of environmental damages under probabilistic causation and asymmetric information is obtained using a linear model.