Soil infiltration, a key process in the terrestrial water cycle, is typically measured pointwise, but is often upscaled by averaging across different soil groups or even texture classes, e.g., when parameterizing water movement in land surface models. We hypothesize that for upscaling, in addition to soil texture, infiltration rates/parameters vary also between different reference soil groups and landuse types. Therefore, we analyzed the between-and within-group variabilities of key infiltration parameters, e.g. saturated hydraulic conductivity (Ks) and final infiltration rate (ic), derived from the Soil Water Infiltration Global (SWIG) database by calculating mutual information and a set of other commonly used statistical measures (e.g., standard deviation) among those classifiers. Results showed that soil texture alone is inadequate to scale up infiltration parameters, leading to lower mutual information and higher standard deviation values of 0.16 and 1.08 for ic, as well as to 0.16 and 3.65 for Ks, respectively. Similarly, landuse also failed to explain the observed variation in infiltration parameters (with mutual information = 0.28 and 0.14 and standard deviation = 1.10 and 4.08 for icand Ks, respectively). In contrast, the World Reference Base soil group was superior to texture and landuse in explaining the observed variability of infiltration parameters, specifically for ic(with higher mutual information and lower standard deviation of 0.52 and 1.10, respectively). The integrated classification of texture, landuse and reference groups resulted in even higher mutual information and lower standard deviation values (with mutual information values of 0.66 and 0.54 for icand Ks, respectively). These results highlight that accounting for the soil classification beyond soil texture should be considered when scaling up the infiltration process.