We describe an algorithm for partitioning the compact parameter space of a model class {f(x^n ; \theta)} into intervals B_i(n) such that the probability P_i(n) = f(B_i(n); \hat{\theta}(x^n)), \hat{theta}(x^n) \in B_i(n), is maximized.  If the central limit theorem holds for the model class, P_i(n) \to P as n \to \infty and m_n/\sqrt{n} \to 1/\sqrt{2}.