Recurrent ponded infiltration experiments on undisturbed samples of coarse sandy loam have revealed a significant flow instability characterized by a decrease in the steady‐state flow rate of the second infiltration run, conducted into wet soil, compared with the first infiltration run, conducted into drier soil. It has been hypothesized that this decrease was caused by air entrapment during the second run, with subsequent blocking of the preferential pathways. In this study, entrapped air distribution and its impact on water flow was studied through a novel combination of magnetic resonance (MR) imaging and numerical modeling. An undisturbed sample of coarse sandy loam was subject to recurrent ponded infiltration while being monitored by MR. Internal structure of the soil sample was visualized by x‐ray computed tomography (CT). A parallel version of the three‐dimensional water flow model based on Richards' equation was used to simulate water flow through the heterogeneous soil sample. Information from the CT was used to describe internal heterogeneity of the soil sample via scaling factors. Magnetic resonance relaxometry imaging data were utilized to derive three‐dimensional maps of entrapped air. These were implemented into a simulation of the second infiltration run as regions of no flow. The MR images were used to assess the water content distribution within the sample. The three‐dimensional model was able to describe measured outflow rates and pressure heads and also to reproduce the heterogeneous distribution of water content within the sample. The results obtained support the assumption that the observed decrease in the outflow rate could be caused by entrapped air in large pores of the soil sample.