The inverse modeling method of Lambot et al. <2002> for estimating the hydraulic properties of partially saturated soils, which was numerically validated, is further tested on laboratory-scale transient flow experiments. The method uses the global multilevel coordinate search algorithm combined sequentially with the local Nelder-Mead simplex algorithm to obtain the inverse of the one-dimensional Richards equation using soil moisture time series measured at three different depths during natural infiltration. Flow experiments were conducted on a homogeneous artificial sand column and three undisturbed soil columns collected from agricultural fields. Three models describing the unsaturated soil hydraulic properties were used and compared: the model of Mualem and van Genuchten, the model of Assouline, and the decoupled van Genuchten--Brooks and Corey combination. The performances of all three models were similar, except for Assouline's model, which provided poorer results in two cases. The inversion method provided relatively good estimates for the water retention curves and also for the saturated conductivity when the moisture range explored was not too small. Water content time series were very well reproduced for the artificial soil and a sandy loam soil, but for two silt loam soils, larger errors were observed. The prediction of the water transfer behavior in the soil columns was poor when flow properties were estimated using directly determined hydraulic properties. The main limiting factor for applying the inversion method, particularly for nonsandy soils, was the characterization of the initial conditions in terms of the pressure head profile. Furthermore, the use of only soil moisture data is essential to enable the hydrogeophysical characterization of soils.