Only dependent variable is considered as random variables in commonly-used linear regression methods, so the regression results will be changed according with the coordinates selection. Enlightened by the method of principal component analysis (PCA), a new unary linear regression which is irrelevant to coordinates is proposed, which is the PCA based method. Compared with conventional least squares method, the new method possesses the advantages of lower deviation error and higher regression accuracy, which is verified by simulation cases and living examples. The simulation case and living example verified new method less system deviation and better regression accuracy than conventional. PCA is numerical solution, the advantage of low calculation amount makes it own a broad application prospect.