The paper presents the concept of soil temperature coefficient, as a ratio of soil temperature in the given point on the area of a basin and soil temperature in the basal point located within the watershed. For modelling the distribution of the soil temperature coefficient depending on selected soil and physiographic parameters, artificial neural networks (ANN) were used. ANN were taught based on empirical data, which covered measurements of soil temperature in 126 points, in the layer of soil at the depth of 0–10 cm, within the area of the Mątny stream basin located in the Gorce mountain range of West Carpathians. The area size of the basin amounts to 1.47 km2. Temperature was measured by means of a TDR device. The soil and physiographic parameters included: slopes, flow direction, clay content, height above sea level, exposition, slope shape, placement on the slope, land-use, and hydrologic group. Parameters were generated using DEM of 5m spatial resolution and soil maps, using the ArcGIS program. The MLP 10-8-1 model proved to be the best fitted neural network, with 8 neurons in the hidden layer. The quality parameters were satisfactory. For the learning set, the quality parameter amounted to 0.805; for the testing set, 0.894; and for the validating set, 0.820. Global sensitivity analysis facilitated the assessment of percentage shares, contributing to the soil temperature ratio. Land use (25.0%) and exposition (20.5%) had the highest impact on of the aforementioned ratio, while the placement on the slope and flow direction had the lowest impact.