The Mociços mine is a Cu-rich quartz-carbonate vein-type deposit hosted in the Palaeozoic detrital metasediments of the Ossa-Morena Zone (SW Portugal), exploited during the beginning of the XX century. A soil geochemistry campaign was carried out, with a portable X-Ray Fluorescence, in order to estimate metal distribution throughout the Mociços mine, followed by summary statistics, interpolation mapping and different machine learning methodological studies (PCA, hclust and k-means) of the obtained data set. Overall results showed that soil geochemistry in the Mociços mine is spatially variable, in which three main populations were recognized (distinguished by all machine learning methods, but better defined by PCA and k-means) and correlated with geological features: i) the mineralized vein, gossan structures, mine tailings and felsic rocks, ii) the geochemical background and iii) a chemically contrasting (Zn higher than regional average) zone. This methodological approach is plausible with typical metal accumulation-dispersion mechanisms in these environments and confirms the relevance of multivariate geochemistry studies for mineral exploration purposes, and of machine learning methods applied to mineral exploration.