AbstractThe development of techniques for fitting non‐parametric smooth curves has resulted in less restrictive regression models. We discuss the ideas underlying such smoothing algorithms, develop their application to epidemiologic studies and address specific issues, such as coping with correlated errors. An example illustrates a particular smoothing approach, as applied to pulmonary function data. The method provides new insight into the effect of smoking on pulmonary function. The discussion offers some qualitative comparisons between smoothing methods and conventional linear models.