This paper presents the stability theorem of cellular neural networks(CNN) with a multi-step output function. Using the different steps of the multi-step output function to store different values of pixels
the CNN associative memory of gray-scale patterns is realized.