This paper deals with the estimation of the theoretical and practical performance efficiency of a subband coding system for still images using memoryless quantizers optimized to entropy and mean-square error constraints. An approach with the concept of rate-distortion theory has allowed to derive a function in special form

which gives the performance gain in terms of signal-to-noise ratio and redundancy reduction. The theoretical upper bounds on entropy coded bit rate reduction were found to be 1.10; 1.51 and 2.10 bits per pixel for M = 4

7 and 16 subbands

respectively. Some simulation results concerning practical performance are presented and interpreted on test images.