A series of tests is carried out using a purpose-built set-up to model compaction grouting around a pile shaft. The tests are performed on Hong Kong completely decomposed granite placed in a Rowe cell, with cement grout directly injected at the bottom of the cell from a pressure interface chamber. The soil specimens are tested under different initial stress and density conditions, either normally consolidated or overconsolidated. Different grouting pressures are used. The results indicate that a target volume of grout can be injected if the grout injection pressure is larger than the total stress by a magnitude at least equal to the effective stress. There is a tendency for the grout to propagate between the pile and the soil, with a residual densification of the soil after drainage. In loose soil, grouting has the effect of reducing the effective stress in the soil to zero, resulting in a homogeneous layer of grout between the shaft and soil. In denser specimens, fractures can occur parallel to the direction of minimum principal stress. A noticeable decrease in void ratio is achieved after grouting, due to both grout injection and drainage.