Proper orthogonal decomposition (POD) is shown to be a statistical operation that identifies the main characteristics of chaotic flows and separates them into a few modes. The dynamic chaotic flow is obtained from two-dimensional (2D) computational fluid dynamics simulations, for different Reynolds numbers, of a confined impinging jets mixer. POD enables reconstruction of the dynamic flow from a few modes that are related to coherent flow structures. The POD flow reconstruction enables a large compression of the flow data set. The decomposition of the flow field into orthogonal modes related to coherent structures provides direct insight into the mixing dynamics and scales which are not accessible from flow dynamics statistic quantities, which were introduced in the context of turbulence and are here applied to chaotic flow.