This paper describes a novel tool called Nimrod/O that allows a user to run an arbitrary computational model as the core of a non-linear optimization process. Nimrod/O allows a user to specify the domain and type of parameters to the model, and also a specification of which output variable is to be minimized or maximized. Accordingly, a user can formulate a question like: “what parameter settings will minimize the model output?”. Nimrod/O currently employs a number of built-in optimization algorithms, namely BFGS, Simplex, Divide and Conquer and Simulated Annealing. Jobs can be executed on a variety of platforms, including distributed clusters and Computational Grid resources. The paper demonstrates the utility of the system with a number of case studies.