A gray scale image storage system applying Complex-Valued Associative Memory (CAM) combined with a 2-dimensional discrete Fourier transform (2-D DFT) process is proposed. In the proposed system, input images are transformed to quantized phase matrices through pre-processing including 2-D DFT then provided to CAM. In addition, the elements corresponding to the high frequency bands in the matrices are omitted in order to reduce the number of neurons of CAM at the degree in which the image degradation is not conspicuous. When a noisy or an imperfect image of a memorized one is input into the system, the corresponding perfect one is output. The image quality memorized into CAM depends on the numbers of both neurons and states of each neuron, respectively. The image quality in changing those numbers and the restoring process of CAM are demonstrated using practical image.