Soil solid thermal conductivity (λs) is critical to model effective soil thermal conductivity (λeff) that is required for engineering design and estimate of soil surface energy flux and soil temperature. However, it is impossible to measure λs because soil is a porous medium and there is no way to compact the soil to a continuous solid state without any pore spaces. The indirect estimation of λs requires saturation of the soils or a complete soil mineralogical information or mineral component such as quartz that has much greater thermal conductivity. Therefore, many approximation approaches of various complexities to predict λs have been proposed. However, few studies have been conducted to assess these models. An extensive review were conducted and returned 20 models to calculate λs. These models were categorized and their performances were assessed with a compiled dataset consisting of 65 soils from five studies. The results showed that the Johansen approach can give satisfactory λs given that quartz content is available (RMSE = 0.60 W m−1 °C−1, NSE = 0.91) and the Tarnawski et al. model suitable for Canadian soils (RMSE = 0.55 W m−1 °C−1, NSE = 0.92). The Côté and Konrad approach that inversely model λs based on the geometric mean model and measured soil thermal conductivity at full saturation (λsat) give accurate λs (RMSE = 0.22 W m−1 °C−1, NSE = 0.99), but cannot be applied to soils without λsat measurement. The other approaches that take use of soil thermal conductivity at dryness (λdry) give unsatisfactory λs. Therefore, a new three-point method (three measurements between λdry and λsat) based on the He et al. model was proposed to predict λs. The results showed this approach provides a reliable method to estimate λs (RMSE = 0.17 W m−1 °C−1, NSE = 0.99) at various textures and water contents without knowledge on mineralogical information. •20 models to estimate solid thermal conductivity (λs) were evaluated•λs can be accurately estimated with the inversing geometric mean model at saturation (one point method)•A new three-point method based on the He et al. 2017 model were proposed to accurately model λs