The ever present possibility of confounding factors creates difficulties in identifying causal effects on the basis of observational data. A large number of approaches to resolve this difficulty have been proposed; see Zaman (2010) for a recent survey. One involves using a "natural experiment," where nature acts like an experimenter in changing the setting of a key variable, allowing us to differentiate between "treatment" and "control" observations. This idea has been used by Hendry and Ericsson (1991), Hoover (2001), and Keane (2010) in rather complex settings. This paper presents an elementary version of this structural approach for detecting causality in the simplest possible setting. The structural method is able to detect contemporaneous causality. We illustrate the uses of this technique on a simulated data set, and also apply it to the export-led growth hypothesis for India and energy-growth data for Shanghai.