Ultra-precision grinding machine is widely used for thinning silicon wafers. The wafer surface quality is considerably affected by volumetric errors at the functional point of the ultra-precision grinding machine. Therefore, a volumetric error model is established based on the homogeneous transformation matrix (HTM) method. The model is, then, optimized by considering Abbe and Bryan errors induced volumetric errors during the error measurements process. A prediction model for the wafer surface topography is established by combining the manufacturing mechanism of the ultra-precision grinding and the optimized volumetric error model of the ultra-precision grinding machine. Volumetric errors are detected using an author’s group developed laser measurement system and a commercial spindle error analyzer. The model enables to predict the topography and total thickness variation (TTV) of the wafer surface. A group of grinding experiments are performed to verify the effectiveness of the proposed models.