With the transparent replica fracture model, two typical LNAPL (light non-aqueous phase liquid) pollutants o-xylene and dodecane) were selected to be flushed by water and the SDS (sodium dodecyl sulfonate) solution. The geometry and distribution of the entrapped LNAPL were obtained directly in a series of experiments. The results showed that the removal percentage of the entrapped LNAPL flushed by water and the SDS solution were in the ranges of 8.3%~12.3% and 65.9%~82.1%, respectively. The cumulative removal percentage were increased with increasing Reynolds number of the flushing fluids. Under the water flushing condition, the entrapped blobs were dispersed into smaller droplets, and the droplet numbers were increased by 1.3~2.2times compared with the initial states. SDS surfactant reduced the interfacial tension between LNAPL and water, thus it can effectively remove the larger entrapped blobs in the fracture. After SDS displacement, only small droplets with a single area of about 1mm² remained. Because of the heterogeneity of variable-aperture fractures, water flush was demonstrated to result in the increase of the effective interface area of ‘LNAPL-water’ in the system, which was disadvantageous to the remediation of LNAPL pollutants. Surfactant enhanced displacement could be a more effective method for LNAPL removal.