1 Proceedings of the 6 th Nordic Symposium on Petrophysics, 15-16 May 2001, Trondheim, Norway Nordic Energy Research Programme Norwegian U. of Science and Technology Forget Better Statistics – Concentrate on Better Sample Selection Patrick Corbett, David Potter, Khalifa Mohammed 2 , and Shujie Liu Department of Petroleum Engineering, Heriot-Watt University, Riccarton, Edinburgh, EH14 4AS, 2 Repsol Oil Operations, Tripoli, Libya Introduction The API’s “Recommended Practices for Core Analysis” makes very little reference to sampling. Core plugs are to be taken at the “selected locations”. The only comments concerning sample volume are for vugs and conglomerates – “it is necessary that sample size be sufficient to include all pebble sizes” (p. 3-5, API, 1998). There are instructions on core orientation: That core plugs should be oriented horizontally and vertically with respect to the core axis or with respect to the bedding planes. The technical procedure for core plugs, probe permeameter and whole cores – including handling and cleaning – is laid out for these industry standard measurements. Similarly, textbooks on Petrophysics do not have sections on sampling (Tiab and Donaldson, 1996). There is therefore no guidance to industry on how samples should be located. There is also nothing to say that core plugs must be taken every foot either, even though this is widely held to be standard industry practice. In this paper, we review the statistical, petrophysical and geological issues for sampling. We propose a series of considerations and are working on a fully recommended procedure. Sampling criteria A set of sampling criteria can be developed by considering the various issues in the various technical interest groups. We consider, in our approach, that the ultimate use of the core data is to sample heterogeneous rocks to get average properties and to build geological models for computer simulation. Other uses of core data are not considered in this short paper. Statistical issues Measurement usefulness: For use in statistical models – this is where much data ends up these days – the key issues are the statistical support and stationarity. Support refers to the appropriate volume of a sample to capture local homogeneity, and stationarity refers to the lateral variation in statistical parameters (averages and variances) associated with that support volume (see Corbett et al., 1999 for further discussion). These issues are well illustrated at the laboratory scale (Figure 1) by comparing probe permeameter data used in clastics (where the support volume is appropriate) with those in carbonates (where there is inappropriate volume support). In this case, either of the measurements in the sandstones could be used in modelling and upscaling, but neither of the measurements are usable in the carbonate – except in the limited intervals of oomoldic grainstone (between 9 and 9.5 cm) where the measurements are consistent. Upscaled flow experiments on cubes support the conclusions that can be made simply by observing the differences between the series of apparent permeability measurements at different scales (Corbett et al, 1999). This issue is partly addressed in the recommended practices with respect to