1: \begin{abstract}

2: Bernstein--von Mises theorems for nonparametric Bayes priors in the

3: Gaussian white noise model are proved. It is demonstrated how such

4: results justify Bayes methods as efficient frequentist inference

5: procedures in a variety of concrete nonparametric problems.

6: Particularly Bayesian credible sets are constructed that have

7: asymptotically exact $1-\alpha$ frequentist coverage level and whose

8: $L^2$-diameter shrinks at the minimax rate of convergence (within

9: logarithmic factors) over H\"older balls. Other applications include

10: general classes of linear and nonlinear functionals and credible bands

11: for auto-convolutions. The assumptions cover nonconjugate product

12: priors defined on general orthonormal bases of $L^2$ satisfying weak

13: conditions.

14: \end{abstract}