The fluid motion inside the engine cylinder is transient, three-dimensional and highly turbulent. It is also well known that cycle-to-cycle flow variations have a considerable influence on cycle-to-cycle combustion variations. Laser-based diagnostic techniques, for example, particle image velocimetry (PIV) or molecular tagging velocimetry, can be used to measure two or three components of the velocity field simultaneously at multiple locations over a plane. The use of proper orthogonal decomposition (POD) allows quantification of cycle-to-cycle flow variations, as demonstrated using PIV data . In the present work, POD is used to explore the cycle-to-cycle flow variations utilizing molecular tagging velocimetry data. The instantaneous velocity fields were obtained over a swirl measurement plane when engine was operated at 1500 rpm and 2500 rpm. The instantaneous flow fields were decomposed into three parts, namely, mean part, coherent part and turbulent part, using triple decomposition approach. The evolution of relative energy content, considering all three parts, was studied at different crank angle positions during intake and compression strokes. Results showed that the mean part is highly correlated, and represents to the bulk flow of the instantaneous velocity field. The coherent part contained about 90% of the total fluctuating kinetic energy, whereas turbulent part contained about 10% of the total fluctuating energy. Hence cycle-to-cycle flow variations are primarily due to the coherent part of the instantaneous velocity field. Concerning different engine speeds, it was found that the fraction of total kinetic energy contained by the mean part was relatively higher for the engine speed of 2500 rpm than that of 1500 rpm.