Amplitude balancing is an important part of the 4D seismic cross-equalization process, which aims to suppress processing and acquisition differences between time-lapse 3D seismic surveys. The matched-filter approach to cross-equalization estimates an amplitude correction that minimizes the norm of the difference section. However, this correction will be biased by the presence of noise, and the amplitude of coherent events will not be equalized correctly. Similarly, a normalization scheme that is based on equalizing the energy in the two surveys, implicitly assumes that their random noise levels are equal. We illustrate the problem with a synthetic example, and present a method that correctly scales the amplitudes based on the relative signal-to-noise levels of each dataset.