Northern Yucatán is a dry tropical area where limestone karst terrain supports a subdeciduous forest that is critical for sustaining the local economy of Mayan people. The 5‐ to 10‐m‐deep vadose zone is characterized by shallow soils (<30 cm thick) with frequent rock outcrops overlying limestone bedrock, which contains the aquifer. This limestone has two important characteristics: (i) lithologic properties that change with depth and (ii) numerous dissolution cavities ranging from small pores to caves, some of them filled with soil (soil pockets). Due to soil shallowness, differences in rock properties are relevant because they restrict or favor root growth; in particular, caves, crevices, and soil pockets allow preferential movement of soil, water, and roots. These features may be important for water and nutrient availability, especially for tree species, but they have not been investigated within an ecological context. We studied the capability of ground‐penetrating radar (GPR) for identifying various limestone layers—the laja (0.3–2.5 m), the sascab (2.5–5 m), and the coquina (5–9 m)—and dissolution cavities. Research was conducted in a limestone quarry so that radar records could be compared with karst features revealed as blasting and rock extraction advanced. Radar records obtained using a 200‐MHz antenna were generally of good interpretative quality; observation depths ranged from 3 to 5 m. In areas that lacked a soil mantle, the interface between the laja and the sascab was clearly identified by GPR at about 2 m. However, the sascab–coquina, and coquina–aquifer boundaries could not be identified. Areas of deep soils (>1 m), microrelief (mounds and plains), and large soil pockets within the limestone matrix were also identified. Main sources of GPR signal attenuation were suspected to be the higher clay and water contents of soil material contained in underlying cavities. As a noninvasive tool, GPR can help to determine properties of the limestone and its dissolution features that are critical to vadose zone–forest interactions.