Flow velocity distribution can be measured by means of a twin-plane tomography system and cross-correlation techniques. However, the use of these techniques based on finding the argument that maximises an estimated cross-correlation function (CCF) of corresponding pixels from two planes, in practice is limited by their low accuracy in reconstructed concentration. Presented is a method permitting the time delay to be measured with higher accuracy. The method is based on utilising the phase of the cross-spectral density of the CCF to estimate a time delay. The method is found to be more promising than the classical correlation method.