Philippe call “multi-source Digital Soil Mapping” a DSM approach that would use together different soil inputs (laboratory analysis, qualitative soil observations, soil sensing data, soil maps, farmers expert saying, existing DSM products, etc.), each of them having different and possibly uncomplete spatial coverages of the study area. This soil data configuration fits better to the real life than the simplified soil data configuration that is most often considered in the literature, i.e. a spatial sampling of measured soil profiles covering more or less evenly the study area with sometimes an exhaustive soil map. Co-kriging could be envisaged as a solution for dealing with multi-source input data but it is hard to use with more than two sources and cannot address efficiently the uncompletedness of the spatial coverages. We already evoked this challenge in the first DSM Global Workshop held in Montpellier in 2004, suggesting that “Spatial Soil Inference Systems” could be the way forward. As nothing has happened since this event, I try to put again this challenge on the agenda. We should revisit it fifteen years after with our increasing experience in DSM … and computing power.