This paper describes the behavior of a low-cost seismic isolator comprising a grout-filled tennis ball rolling on concrete plates. The isolator could be used for isolating lightweight structures. Initially, the axial response of the system was characterized. Subsequently, full-scale isolators were tested in combined compression and lateral cyclic loading. Parameters of investigation were the testing velocity (frequency), the bearing load, the degradation due to consecutive loading, the bearing temperature, the specimen-to-specimen variability, and the engagement of the displacement restrainer. Finally, a three-dimensional finite element model was developed to model in detail the response of the isolator and to explore the influence of the rubber thickness. This is the first study to characterize the above effects and to propose a finite element model of these isolators. Results showed that the lateral cyclic response of the isolators is bilinear and can be approximated by rigid-body equations, whereas the values of the rolling friction coefficient and the yield displacement are suitable for seismic isolation applications. The specimen-to-specimen variability was minimal. Unlike sliding isolation systems, an increased bearing compressive load leads to a higher rolling friction coefficient. Analytical equations are offered to describe this effect. The rolling friction coefficient does not depend on velocity (frequency) or temperature, and the isolators did not deteriorate under consecutive loading. The proposed displacement restrainer effectively limits the motion of the isolator. The developed finite element model accurately captures the experimental response. It was used to study the influence of the thickness of the rubber layer on the lateral response, to conclude that it is minimal, for rubber layers ranging from 3.4 to 7 mm.