A prediction model of the cycle-to-cycle variation of the in-cylinder flow in IC engines which employs the time averaged k-ε turbulence model is proposed. The concept is based on an assumption that the power spectrum of the cycle-to-cycle variation can be deduced from the power spectra of both the mean velocity and turbulence intensity. To validate this model, in-cylinder velocity measurement in a transparent cylinder engine with a 2-valve cylinder-head is made using an LDV system. Comparisons of in-cylinder flow fields between the calculation and measurement show a good agreement in the cycle-to-cycle variation as well as the turbulence intensity. Finally, this model is applied to three kinds of flow fields to examine how the cycle-to-cycle variation may be effected. As a result, it is found that the swirl flow is effective to reduce the cycle-to-cycle variation, while the tumbling flow enhances the turbulence generation around the compression TDC.