Engine in-cylinder flow varies from cycle to cycle, which contributes to variation of the mixing and combustion processes between fuel and air. Such flow field cyclic variability at the macroscopic scale is distinct from random fluctuations at the microscopic scale about the ensemble mean velocity field due to turbulence. At the extreme, the mean velocity field may bear no resemblance to any instantaneous flow field within the ensemble. Rather, these instantaneous fields may appear multimodal. Yet previous attempts to define and identify the flow modes were either qualitative (by visual inspection), or based on strict point-by-point velocity difference between two flow fields. The former approach is clearly subjective; the latter does not accommodate translational and rotational variations of in-cylinder flow patterns relative to a flow mode. Such spatial variations, in location and orientation, of the flow patterns can be quantified by the technique of complex moment normalization. The algebraic properties of complex moments are also intimately related to the geometric and physical properties of two-dimensional/two-component flow fields. In this paper, we take the normalized moments as flow field attributes for further cluster analysis. This analysis approach is demonstrated using a set of in-cylinder flow fields obtained by high-speed particle image velocimetry on a swirl plane of a research optical engine operating under low intake swirl setting. The resulting classification of the flow fields into several clusters (flow modes) are discussed, and the potential and limitations of the analysis approach are appraised.