This paper makes three points relevant to the application of the precautionary principle to the regulation of GMOs. i) The unavoidable arbitrariness in the application of the precautionary principle reflects a deeper epistemological problem affecting scientific analyses of sustainability. This requires understanding the difference between the concepts of "risk", "uncertainty" and "ignorance". ii) When dealing with evolutionary processes it is impossible to ban uncertainty and ignorance from scientific models. Hence, traditional risk analysis (probability distributions and exact numerical models) becomes powerless. Other forms of scientific knowledge (general principles or metaphors) may be useful alternatives. iii) The existence of ecological hazards per se should not be used as a reason to stop innovations altogether. However, the precautionary principle entails that scientists move away from the concept of "substantive rationality" (trying to indicate to society optimal solutions) to that of "procedural rationality" (trying to help society to find "satisficing" solutions).