A significant reduction in the costs associated with contamination assessments can be achieved if traditional soil sampling for contaminated-site characterization is complemented by real-time sampling using proximal soil sensors. Real-time sampling using a portable X-ray fluorescence (pXRF) device is a cheap and fast sampling method to provide more data and reduce the time needed to map soil contamination. The main disadvantage of using pXRF is the degree of uncertainty of these in situ measurements due to the technology’s indirect nature, and its sensitivity to soil heterogeneity and soil moisture content. This study evaluates the potential of using both pXRF and traditional soil sampling measurements to accurately map soil contamination due to the presence of heavy metals. The approach proposed uses geostatistical sequential simulation with local probability distributions to characterize and integrate pXRF uncertainty at each sampling location. The resulting maps agree with the contamination map obtained using traditional laboratory data only, in terms of mapping accuracy and extent of contaminated areas. This study shows that with few collocated pXRF and laboratory analytical data it is possible to identify contaminated areas accurately, thus providing a cost-effective solution to work with pXRF data directly.