A brief description is given of a model representing the in-core behavior of a single Advance Gas-Cooled Reactor (AGR) fuel stringer, developed specifically for fuel performance optimization studies. The difficulties encountered in tackling such a problem using standard library packages are outlined and the performance of the only sitable NAG routine is compared with that of a specially written Metropolis ('Simulated Annealing') algorithm program. It is concluded that in this application the stochastic Metropolis algorithm has several distinct advantages over the deterministic NAG routine.