Gas Turbine development tests are expensive, time-consuming, and even after completion of numerous runs of experiments, often produce ambiguous results on key performance metrics. Measurements of efficiency and power output are influenced by ambient conditions, and despite attempts to correct for such influences, the uncertainties introduced by these variable conditions are often much larger than the differences in performance between machines that the tests are designed to detect. Furthermore, when an experimental run needs to be undertaken under specific, uncontrollable ambient conditions, it is currently unclear if there is an optimal set of experiments that should be undertaken to maximise the value to the engineers of these expensive tests. It is also sometimes unclear how the experimental results should be best be used to compliment the results of much cheaper computational simulations beyond the use of simple correlations. In response to these issues, a new generalised low-order model of the gas turbine with a high degree of parametrization is introduced. The novel approach to the problem of effective development tests is to impose a probability distribution on the parameter space, and to update this distribution in a Bayesian manner in response to a dataset composed of experimental and computational results of variable fidelities. The posterior distribution of the parameter space expresses the engineers updated beliefs about the underlying performance of the turbine, and can then be sampled to predict performance in untested regions in a probabilistic manner. In addition, the parameter space can also be used to calculate an expected utility of a future experiment, enabling a more cost effective-use of experimentalists’ resources.