This paper presents an approach to the evaluation of reservoir models using transient pressure data. Braided fluvial sandstones exposed in cliffs in SW England were studied as the surface equivalent of the Triassic Sherwood Sandstone, a reservoir unit at the nearby Wytch Farm oilfield. Three reservoir models were built; each used a different modelling approach ranging in complexity from stochastic pixel-based modelling using commercially available software, to a spreadsheet random number generator. In order to test these models, numerical well test simulations were conducted using sector models extracted from the geological models constructed. The simulation results were then evaluated against the actual well test data in order to find the model which best represented the field geology. Two wells at Wytch Farm field were studied. The results suggested that for one of the sampled wells, the model built using the spreadsheet random number generator gave the best match to the well test data. In the well, the permeability from the test interpretation matched the geometric average permeability. This average is the "correct" upscaled permeability for a random system, and this was consistent with the random nature of the geological model. For the second well investigated, a more complex "channel object" model appeared to fit the dynamic data better. All the models were built with stationary properties. However, the well test data suggested that some parts of the field have different statistical properties and hence show non-stationarity. These differences would have to be built into the model representing the local geology. This study presents a workflow that is not yet considered standard in the oil industry, and the use of dynamic data to evaluate geological models requires further development. The study highlights the fact that the comparison or matching of results from reservoir models and well-test analyses is not always straightforward in that different models may match different wells. The study emphasises the need for integrated analyses of geological and engineering data. The methods and procedures presented are intended to form a feedback loop which can be used to evaluate the representivity of a geological model.