Principal curves were introduced to formalize the notion of ''a curve
passing through the middle of a dataset.'' Vaguely speaking, a curve i
s said to pass through the middle of a dataset if every point on the c
urve is the average of the observations projecting onto it. This idea
can be made precise by defining principal curves for probability densi
ties. In this paper we study principal curves in the plane. Like linea
r principal components, principal curves are critical points of the ex
pected squared distance from the data. However, the largest and smalle
st principal components are extrema of the distance, whereas all princ
ipal curves are saddle points. This explains why cross-validation does
not appear to be a viable method for choosing the complexity of princ
ipal curve estimates.