Strong cycle-to-cycle variations of fuel spray are observed due to the highly transient in-cylinder airflow in spark-ignition direct-injection (SIDI) engine. The spray structure comparison based on ensemble-averaged image may be misleading sometimes because the spray images for the same engine running condition could be different from cycle to cycle. Also, the visual comparison of spray images from many cycles is only qualitative and very time-consuming. Therefore, the present paper provides a novel approach to make quantitative comparison of spray structures from different engine conditions, or comparison between experiment and simulation (such as large eddy simulation, LES). The methodology is based on the proper orthogonal decomposition (POD), which has been utilized for in-cylinder turbulent flow research for over a decade.The spray images at two distinct intake air conditions recorded inside a single-cylinder optical engine were used to demonstrate this novel POD analysis approach to distinguish the spray variations. One intake air condition was set at low swirl flow (swirl ratio of 0.55), and the other was set at high swirl flow (swirl ratio of 5.68). Spray images of both conditions were merged into a single dataset for the POD analysis. After conditional sampling on the POD coefficients, it can be found that there was a distinct separation of POD coefficients for two swirl conditions. The differences in the POD coefficients together with their corresponding POD mode patterns clearly distinguish the variation between two air flow conditions. The range of the coefficient distribution can be used to quantify the spray structure variations, which is believed to be caused by the difference in the airflows at both swirl levels. In summary, the present technique is anticipated to providing a useful tool for researchers who are looking for methods to make quantitative and systematic comparison of spray structure variations in engine flows.