Engineers occupied in the development of hardware coder-decoder pairs (codecs) need an objective performance measure. Such a measure is the mean-square error (MSE). For complicated signals, such as speech signals, the error signal e(t) = Si(t) − s0(t) is hard to obtain. The reason for this is that each codec has a nonzero delay and a nonunity gain. This makes a straightforward subtraction of the output from the input signal impossible. This paper describes an accurate microprocessor-based measuring system applicable for the evaluation of speech codecs. In this system, the nonzero delay and the nonunity gain of the codec under test are compensated. Thus the error signal can be obtained by means of a differential amplifier. During a predefined period, the input and error signal are measured and the mean-square values of both are calculated. The ratio of these values, expressed in decibels is presented as the result of the measurement. Any input signal can be used, but since the system is intended for codecs of communication quality, the frequency range is limited to 3400 Hz. Signal-to-noise ratios (SNR's) of up to 70 dB can be measured.