There is considerable interest in watershed-based pollution water quality protection but the approach can be highly information intensive (USEPA 2004, NRC 2000). This study examines the value of different types and levels of information for water quality management in the Conestoga watershed. For this estimation, a Monte Carlo procedure is used to construct the posterior expected value. Then, an Evolutionary Optimization Strategy with Covariance Matrix Adaptation (CMA-ES) is used to compute the expected value of optimized resources allocations given posterior information structures for specific sample sizes. This posterior optimization is nested within a second Monte Carlo simulation that computes the preposterior expectation (a nested Monte Carlo procedure). Thus, this paper provides some insight about the relative values of these alternative types of information for controlling water pollution from agriculture, and the gains from more intensive sampling.