Insulin pump controllers seek to alleviate the chronic suffering caused by diabetes that affects over 6% of the world population. The design of control laws for insulin pump controllers has been well studied. However, the parameters involved in the control law are difficult to synthesize. Traditionally, ad hoc approaches using animal models and random sampling have been used to construct these parameters. We suggest a synthesis algorithm that uses Bayesian statistical model validation to reduce the number of simulations needed. We apply this algorithm to the problem of insulin pump controller synthesis using in silico simulation of the glucose-insulin metabolism model.