Soil‐probe contact resistance and finite radius and heat capacity of the heat pulse (HP) probe produce significant errors in thermal property estimates. In this study, we demonstrated that estimating soil thermal properties from late‐time data of the temperature change‐by‐time (ΔT(t)) curve reduces these errors effectively. The weighted nonlinear curve fitting method was applied to estimate soil thermal properties following the pulsed infinite line source (PILS) theory using ΔT(t) data from the complete (PILS‐Complete), peak‐time (PILS‐Peak), and late‐time (PILS‐Late) ranges. Three experiments on specific heat of soil solids (cs), soil thermal properties, and soil water content (θHP) were conducted to examine the performance of these approaches. The results showed that the PILS‐Complete and PILS‐Peak methods overestimated cs by 16.6% and 13.0% respectively, and the error from the PILS‐Late method reduced to 3.2%. Soil thermal conductivity measurements from the PILS‐Late method agreed well with those from the identical‐cylindrical‐perfect‐conductors theory and with the estimates from the heat flux plate data. The PILS‐Late method also effectively reduced the overestimation of soil heat capacity and underestimation of soil thermal diffusivity. In comparing to the PILS‐Complete method, the PILS‐Late method reduced the root mean square error (RMSE) of θHP from 0.039 to 0.021 m3 m−3 on a sand soil, and from 0.032 to 0.018 m3 m−3 on a clay loam soil. Thus, using late‐time data improved the accuracy of HP method for measuring soil thermal properties.