Digital soil mapping (DSM) is commonly conducted using input soil attributes derived from laboratory analyses of geographically referenced samples. Field observations are often abundant and can offer a dense source of soil data that has the potential to enhance DSM predictions. However, they are not widely used due to to concerns about subjectivity and data quality. This study investigates the usefulness of hand-feel soil texture (HFST) data for DSM. We processed HFST data obtained from forest soils in France from two inventory campaigns: (i) HFST determination from systematic 1 km2 grid observations in France utilizing a specialized soil texture triangle, and (ii) HFST observations from soil survey samples, using a different texture triangle. Both sets of HFST data were used as input soil variables, with the same covariates, for predicting topsoil texture In a sizable, forested area through a DSM method. By employing independent sampling and laboratory soil analyses in selected areas, we uncovered measurement bias in one of the datasets. However, intriguingly, these biased observations identified subtle yet highly specific and unexpected patterns of sands in terraces due to alluvial deposits along small rivers. Thus, field soil observations, even if they are biased, should not be dismissed solely based on their overall predictive performance. It is essential to carefully examine predicted maps and covariates to determine whether patterns may have pedological and/or lithological origins and if they are pertinent for enhancing DSM predictions, enhancing soil process understanding, and meeting the requirements of end users. Numerous HFST are available worldwide, these datasets are usually disregarded for DSM. Here we contend that efforts should be put in recovering these data, and their potential for enhancing DSM and deepening our understanding of soil processes.