Conventionally, performance of Image Coder is evaluated by SNR of reconstructed pictures. However, SNR is not a good criterion. For instance, blocking artifacts are more perceptible than a random noise and deteriorate picture quality. In this paper, we propose a new measure of picture quality considering not only a random noise but also a correlated noise, Five deterioration factors are considered ; fl : weighted mean square error (WMSE), f2 : WMSE lager than visual threshold, f3 : changes of errors between successive pixels at the block boundaries, f4 : autocorrelation of errors, f5 : measure of errors around the contours. Careful assessment tests are carried out and a new measure is obtained by utilizing multiple regression analysis and principal component analysis. The individual principal component obtained by principal component analysis represents characteristics of error clearly, and the dependence of measure on the type of image is greatly improved. Simulation results show that this measure well approximate the Mean Opinion Scores.