This paper proposes an iterative face occlusion removal algorithm based on accumulative error compensation and fuzzy principal component analysis (FPCA). The originality of this work is two folds. First, instead of successive error, normalized accumulated absolute error was used as an image fusion weight in recursive error compensation. Second, gappy PCA with bi-value mask was extended to fuzzy PCA with continuous mask between 0~ 1. The value of the fuzzy mask vector is also defined by normalized accumulated error, which indicates the probability of being occluded of face region. Experimental results shown that our new reconstruction algorithm could remove unknown occlusion with various shape effectively, and outperform classical iterative PCA based algorithm.