Correlation in sand-rich deep-water clastic reservoirs using sedimentary characteristics is examined using well-exposed outcrop data. Individual bed thickness, or the sedimentary characteristics of single beds, and packages of beds when viewed in one dimension are found to give an inadequate basis for interwell correlation over distances greater than 100 m. Offset stacking of beds is the main control over bed thickness variations with scouring playing a minor role. Composite units are identified that can be traced over kilometer-scale distances. These units are potentially identifiable on seismic data and are valuable for interwell-scale reservoir modeling. Pinch-out of sand units is characterized in two end-member styles, onlap and infill. Onlap and infill indicate the influence of paleotopography on deposition, reflecting deposition onto low and high relief, respectively. Onlap is characterized by gradual deterioration of reservoir quality toward the pinch-out, whereas infill maintains good reservoir quality without significant deterioration close to the pinch-out. Prediction of the proximity of 1-D (one-dimensional) sections to termination of reservoir units at a pinch-out is problematic for both styles. Although 1-D sections provide invaluable data for reservoir characterization prediction of 3-D (three-dimensional) sand body geometry from borehole sections is problematic. Commonly cited channelized and lobate sandstone body geometry does not have diagnostic sedimentary features and recognition of alternative sand body geometry is needed. The difficulty in selecting key, small-scale sedimentary features that, in turn, are diagnostic of larger scale geometric features limits their value in the up-scaling of reservoir characteristics. A new approach to reservoir modeling of sand-rich deep-water clastic reservoirs is proposed that involves down-scaling from mapping of seismically detectable, interwell-scale packages of beds and the paleotopography on which they lie. Once these features are constrained smaller scale characteristics are modeled.