Textural and mineralogical properties of the sediment, facies characteristics (both derived from well data), stacking patterns (derived from well and seismic data) and plan-view shapes (derived from seismic data) are all essential components when reconstructing reservoir geometries from subsurface data sets. Without the availability of all these sources of information, reservoir predictions can potentially deviate significantly from the true geometries. In particular, inferences about internal sandbody geometry from plan-view considerations (isopach maps and seismic attribute maps) may lead to erroneous conclusions without knowledge of facies and textural properties of the sediment. Analysis of well-exposed sedimentary systems has the potential to establish links between internal facies characteristics and large-scale geometry and to improve models coupling information at these variable scales. The Palaeogene Battfjellet Formation, Svalbard, has been studied with the aim to unravel internal characteristics and external form by investigating sediment properties, facies and stacking patterns. The formation shows a combination of texturally and mineralogically very immature sediments, a predominance of wave-generated or wave-induced sedimentary structures and a stacking pattern of highly variable numbers of parasequences at localities few kilometres apart that best can be explained in terms of small shifting deltaic lobes that produced a complex pattern of overlapping sandbodies. The strong evidence of wave action in the receiving basin could in itself indicate strike-extensive sandstone bodies; however, the complex sandbody arrangement and the immaturity of the sediments preclude such an interpretation. Traditional facies models coupling plan-view geometries with internal facies characteristics (such as the coupling of strike extensive barrier systems with wave-dominated sedimentary structures or the coupling of elongated fluvial-dominated deltas with offshore-directed current-generated structures) are much too simple and may lead to erroneous interpretations if the complexity revealed by all sources of data is not appreciated.