Forty field soil samples, from nine Canadian Provinces, were laboratory tested for their ability to conduct heat, i.e., thermal conductivity ( $$\lambda $$ ), using a non-stationary probe technique. The measurements were carried out on moderately compacted samples at room temperature, and over a full range of degree of saturation ( $$S_{\mathrm{r}})$$ ranging from dryness to full saturation. The analysis of $$\lambda $$ data revealed a strong nonlinear variation of $$\lambda $$ versus $$S_{\mathrm{r}}$$ that can be sub-analyzed in four $$S_{\mathrm{r}}$$ -based domains, i.e., residual, transitory meniscus, micro/macro-porous capillary, and superfluous. In the residual domain ( $$0 < S_{\mathrm{r}}< S_{{\hbox {r-cr}}})$$ , a very small $$\lambda $$ variation is observed. In the transitory meniscus domain ( $$S_{{\hbox {r-cr}}} < S_{\mathrm{r}} < S_{{\hbox {r-}\mathrm{PWP}}})$$ , for the majority of soils, a sharp $$\lambda $$ increase was observed. In the micro/macro-porous capillary domain ( $$S_{{\hbox {r-}\mathrm{PWP}}} < S_{\mathrm{r}} < S_{{\hbox {r-}\mathrm{FC}}})$$ , a moderate $$\lambda $$ increase was noted. Finally, in the superfluous domain ( $$S_{{\hbox {r-}\mathrm{FC}}} < S_{\mathrm{r}} < S_{{\hbox {r-sat}}})$$ , a very slight $$\lambda $$ increase was generally noted. On average, the highest $$\lambda $$ values (from 1.9 $$\hbox {W}{\cdot }\hbox {m}^{-1}{\cdot } \hbox {K}^{-1}$$ to 3.2 $$\hbox {W} {\cdot }\hbox {m}^{-1}{\cdot } \hbox {K}^{-1})$$ were obtained from saturated soil samples with high quartz content, e.g., samples from sites in Nova Scotia and Prince Edward Island. On the other hand, the lowest $$\lambda $$ (from 1.1 $$\hbox {W}{\cdot }\hbox {m}^{-1}{\cdot }\hbox {K}^{-1}$$ to 1.4 $$\hbox {W}{\cdot }\hbox {m}^{-1}{\cdot }\hbox {K}^{-1})$$ was observed from saturated soil samples with lower quartz content, e.g., British Columbia samples. The measured data were used to verify a recently developed series–parallel (S– $${\vert }{\vert }$$ ) model for unsaturated soils. On average, estimates of the S– $${\vert }{\vert }$$ model, with a parallel arrangement of air and water in the third of three conductive paths, were within $${\pm }0.08\,\hbox {W}{\cdot }\hbox {m}^{-1}{\cdot }\hbox {K}^{-1}$$ of experimental data. However, the S– $${\vert }{\vert }$$ model, with a series arrangement of air and water in the third conductive path, showed slightly better estimates when it was applied to fine-textured soils. In addition, there was no strong correlation noted between the performance of the S– $${\vert }{\vert }$$ models and soil quartz content. Consequently, it is recommended to compare estimates from both models when they are applied to experimental data.