Let a nonparametric regression model , where is respons variable, is regression curve that assumed an unrestricted form and contained in Sobolev space . For estimate curve is obtained by minimizing the Penalized Least Square (PLS). In this case given cubic spline polynomial approaching for optimal knots points, by using Generalized Cross Validation (GCV) method, to obtained optimal estimation model for regression curve. This application of cubic spline using bread turnover data from CV DEDE MAKASSAR. Based on analysis obtained four optimal knots on the months 3, 6, 8, and 11 by estimation equation as follows : Keywords : PLS, cubic spline, optimal knots, GCV.