According to the Swedish Environmental Protection Agency, the number of potentially contaminated sites in Sweden is approximately 40 000. The funding for remediation from the Swedish Government alone amounts to $A 60 - 80 M (300 - 400 MSEK) annually for the period 2003 - 2006.Uncertainties will arise when obtaining samples in field, preparing them in laboratories and analysing them. Different procedures give rise to varying errors. Using data for decision-making that does not truly represent the site could be fatal. Misclassification of soil could lead to unnecessary costs or to risks to human health and the environment.The total uncertainty of an environmental sample can be split into analytical uncertainty and sampling uncertainty. The analytical uncertainty is usually known and reported from routine analysis. The size of the sampling uncertainty is less well known. Hence, the question is how big are the sampling uncertainties for contaminated soil and how big is the analytical error compared to the total sampling error?A scrap yard and two landfills within a workshop area were investigated. An excavator dug trial pits and samples were taken with a trowel from the walls of the trial pits. Samples were put into plastic bags and taken off site for preparation and analysis using XRF. Sample preparation included drying and sieving. The samples were homogenised by kneading the sample bags prior to analysis.The sampling and sample handling were performed in an intuitively way without knowledge of sampling theory. However, Pierre Gy's sampling theory was used for subsequent evaluation of the data from the site investigation. The theory defines seven sampling errors, which can be applied to sampling of soil. Some errors have a variance that can be calculated or at least estimated, while other errors can give some idea whether the bias is positive or negative. The total sampling error varied over three orders of magnitudes from sample to sample. The single largest contributor to the large variation in error was the contaminant level. When comparing the analytical error and the total sampling error the ratio varies more than three orders of magnitude.