Health economists frequently wish to estimate causal relationships from observational data, for example, the effect of education on health. Such inference problems can be very challenging, as associations in observational data typically reflect causation in either direction, or common but possibly unobserved causes of both variables. Continuing the example, an observed correlation between health and education does not reveal the causal effect of education on health, as poor health could lead to lower education, and many other variables, such as cognitive and noncognitive aspects of personality, may lead to changes in both health and education. Instrumental variables (IV) estimators are commonly deployed in health economics to attempt to estimate causal relationships when limited to observational data. In this article, the use and limitations of the IV approach are discussed.