For efficient coding of graphic image data, the statistical characteristics for both Korean lettered images and English lettered images are measurpd and co mpared. Also, the measured run length distribution is compared with the run length distribution hased on Markov model. It is shown that the measured white run length distribution is more Bike a negative - power distribution than an exponential distribution . This fact is stronger in the Korean lettered images than is the English lettered images, The performances of four typical run length codes are compared for the same set of graphic data files,, and it is shown that the codes perform better in the Korean ]entered images :hart In Eng]isle lettered images.