In iron-bearing rock provenance researches the definition of what is a source vary from an author to another. Issues might occur when the sources are associated with complex weathering profiles. How may the definition of the source influence the results of provenance studies when different types of iron-bearing rocks are encountered within the same weathering profile? In order to deal with this research question, we used a geochemical dataset previously obtained from the study of the ochre geological collection collected around Diepkloof rock shelter, South Africa. We tested the efficiency of Linear Discriminant Analyses (LDA, supervised analysis) and Principal Component Analysis (PCA, non-supervised analysis) in source discrimination and in the attribution of archaeological pieces to the right source. Different assumptions were made: sources were defined by separating or grouping the different lithologies (shale and ferricrete) encountered at the same location. The results show that grouping the different lithologies allows a good attribution of the archaeological pieces to the source it comes from when elements positively correlated with iron are used in the analysis. Separating the different lithologies do not allow a rate of success in archaeological pieces attribution as high as when the lithologies are grouped together. This result might be explained by a highest number of samples that define a source when the different lithologies are grouped together. Further research should be considered in this direction in order to define if this is a general trend.