This paper is concerned with infrastructure maintenance and repair policies when the managing agency does not value expenses linearly. Such nonlinearity naturally arises in the stochastic world of reliability engineering, manifesting, for instance, as risk aversion: the ideal maintenance and repair policy has a low cost variance, as well as a low average cost. Another type of nonlinear behavior arises when one tries to match future expenditures with an externally determined budget. Dynamic programing techniques are applied to create two algorithms (FindPolicy and EvalPolicy) which are of use in this problem: FindPolicy determines an optimal maintenance policy, while EvalPolicy allows a previously determined policy to be evaluated according to a broad class of measures of effectiveness. These algorithms are applied to a hypothetical bridge facility and to determine and evaluate short-term and long-term maintenance policies. In this example, large reductions in solution variance are attainable with only slight increases in average cost.