This work presents a tiling algorithm to construct multi-layer perceptron networks with perfect performance for all training patterns. Each perceptron distinguishes, as much as possible, a portion of a single class patterns from all the rest patterns. It is a divide and conquer approach. Each layer of such perceptrons encodes different class patterns with different codes, faithful representations. This construction provides resolution of challenging classification of seriously corrupted patterns. To our knowledge, it gives the highest performance for corrupted testing patterns among all existing methods.