Summary Determination of infiltration properties of soils under laboratory conditions necessitates the collection of soil samples in a way that maintains their natural physical properties. Mountain forest soils, containing rock fragments, root systems and a significant amount of organic matter, make it extremely difficult to test their hydraulic conductivity using both laboratory and field methods. A widely used technique of sampling by driving a cylinder into the ground in this type of soils causes damage to their structure resulting from the displacement of root systems and rock fragments as well as reduction of soil porosity. Thus, subsequent results contain an error that is difficult to estimate. The aim of the present research was: (1) to develop a laboratory method for testing the hydraulic conductivity of mountain forest soils, and in particular a method of collection of undisturbed soil samples, (2) to determine the influence of the applied method of collecting samples on the thickening of their peripheral layer and on elimination of increased infiltration at the boundary between the soil medium and the cylinder, (3) to determine the extent of the impact of the irregular shape of a sample on its hydraulic conductivity and (4) to develop an empirical method for determining the actual values of hydraulic conductivity, taking into account the error associated with the flow of water through samples with different shapes. The method of soil sampling consists in gradual formation of a cylindrical soil monolith and filling the free space between the monolith and the tri-cylindrical container with low-pressure assembly foam. This method ensures preservation of the natural physical properties of the examined samples and elimination of errors during the measurement of the hydraulic conductivity, caused by increased infiltration at the boundary between the soil medium and the cylinder. It was shown that the mean error of hydraulic conductivity determination, related to the irregular shape of samples, amounts to 11.57%. The error may be eliminated by the application of conversion coefficients.