It is provided a computer-implemented method for compressing a three-dimensional modeled object, wherein the method comprises providing a boundary representation of the modeled object; determining first data, that describe a mapping from the range of a reference function to the range of another function, and second data, that describe a mapping from the domain of the other function to the domain of the reference function which, when composed with the reference function and the mapping from the range of the reference function to the range of the other function, leads to the same result as applying the other function; and replacing in the geometrical data the other function by the first data, the second data and a pointer to the reference function. Such a method improves the compression of a 3D modeled object.