The rear surface of a freeform progressive addition lens （PAL） is often fitted using multiple surfaces， and curvature changes at the joints are usually large and discontinuous. Initial freeform PALs frequently required local or global optimization to improve their optical properties and surface smoothness. A global fitting method based on the non-uniform rational B-spline curve and least squares curve is proposed to optimize reconstruction of the rear surface of a lens. The algorithm principle of least squares curve fitting is introduced， and the rear surfaces of two sets of lenses with different powers are reconstructed under the same parameters； moreover， their optical properties are simulated， quantified， and evaluated. Experimental results demonstrate that the maximum astigmatism of both lens groups reduces by 0.61 D and 0.45 D， respectively. The astigmatism in the distance area decreases by 0.15 D and 0.16 D， respectively， and the middle channel width （Y = 0 mm） increases by over 1.8 mm. The proposed reconstructed surface improves the rear surface smoothness， gradient channel width， and effectively reduces PAL maximum astigmatism.