Together, the three particle size fractions (PSFs) of clay, silt, and sand are the most fundamental soil properties because the relative abundance influences the physical, chemical, and biological activities in soil. Unfortunately, determining PSFs requires a laboratory method which is time-consuming. One way to add value is to use digital soil mapping, which relies on empirical models, such as multiple linear regression (MLR), to couple ancillary data to PSFs. This approach does not account for the special requirements of compositional data. Here, ancillary data were coupled, via MLR modelling, to additive log-ratio (ALR) or isometric log-ratio (ILR) transformations of the PSFs to meet these requirements. These three approaches (MLR vs. ALR-MLR and ILR-MLR) were evaluated along with the use of different ancillary data that included proximally sensed gamma-ray spectrometry, electromagnetic induction, and elevation data. In addition, how the prediction might be improved was examined using ancillary data that was measured on transects and was compared to data interpolated from transects spaced far apart. Although the ALR-MLR approach did not produce significantly better results, it predicted soil PSFs that summed to 100 and had the advantage of interpreting the ancillary data relative to the original coordinates (i.e. clay, silt, and sand). For the prediction of PSFs at various depths, all ancillary data were useful. Elevation and gamma-ray data were slightly better for topsoil and elevation and electromagnetic (EM) data were better for subsoil prediction. In addition, a smaller transect spacing (26 m) and number of samples (9–16) might be adopted for mapping soil PSFs and soil texture across the study field. The ALR-MLR approach can be applied elsewhere to map the spatial distribution of clay minerals.