This paper proposes a method for simulating real-time haptic interaction with deformable objects. The deformable model consists of regular hexahedrons of a single type. This homogeneity is exploited to improve the efficiency in deformation computations. Model boundaries are approximated using a moving-least-squares function reflecting the deformation results of the hexahedrons. A method for adaptively approximating the model boundaries is presented for efficient collision handling in the haptic loop. The proposed method can simulate a model of 16,481 nodes in less than 1 ms, which is a significant improvement over the previous methods in the literature. Small gap between the model boundary and the hexahedrons can cause errors in the proposed method. Numerical examples considering the characteristics of human tissues show that the errors are less than just-noticeable difference of human.