Bayesian nonparametrics is not an oxymoron, but refers to using a probability distribution on probability distributions, not just the parameters of a distribution, as a prior. These models promise a reasonable approach to robustness, and have other strengths. But they do have mathematical difficulties and it may be difficult to incorporate prior information in a reasonable manner. No definitive answers will be proposed. The organizer knows of none. However, it is hoped that there will be a healthy discussion of the possibilities of such models in drug development and regulatory research.