A relatively new Multiple Point Statistics algorithm, Direct Sampling (DS), has been tested using a turbidite lobe succession as an example. A seismically derived probability cube was used as auxiliary data. The results show that the combination of low resolution auxiliary data (reservoir is seismically tuned) and a high resolution training image produces geologically realistic reservoir facies geometries. It is shown that it is convenient to use the reservoir grid as a training image, which is an efficient way in order to achieve a reliable correlation between facies in the training image and auxiliary data. Since the training image is highly non-stationary, only the realisations obtained when applying a high weight (>0.4) of the auxiliary variable can be appropriately used. Simulations produced by the IMPALA algorithm are less convincing, which leads to the conclusion that it is recommendable to use DS in similar cases.