Freeform optics, offering high degrees of design freeform to control light propagation, have already been widely applied in various photoelectric equipment. The form quality of those optics is crucial to their opto-electronics functionalities, which requires to be measured accurately. The deflectometry is a promising technology to test the complex freeform surfaces. In general, there is a designed surface model for the monoscopic deflectometry to estimate the positions of whole measured points to solve the issue of height-slope ambiguity. However, the unknown or inaccurate surface model can induce errors into the measured normal, thereby decreasing the measurement precision. In this paper, without relying on the known surface model, the proposed method iteratively optimizes a sphere model to describe the measured surface by changing the spherical radius. In order to reduce the global error, the space-variant spheres are optimized, respectively, to estimate the whole-aperture surface coordinate. With the help of the iteration surface reconstruction process, the optimal number of the space-variant spheres is achieved to meantime obtain the final reconstructed surface. Compared to the measurements by using the plane model, the form accuracy can be improved by three times. Experiments demonstrate that the proposed method can successfully reconstruct the complex surfaces without the need of a known surface model, which can greatly improve the measuring flexibility and measurement accuracy.