Previous studies of corneal and keratometric variation used statistica
l methods that were not entirely satisfactory. For the first time, pro
per multivariate statistical methods are applied to evaluate short-ter
m keratometric variation in human eyes. Keratometric variation is repr
esented graphically by means of stereo-pair scatter plots, ellipsoidal
confidence regions for mean dioptric power, and meridional profiles o
f variation. There is great variability in the keratometric variation
displayed by different subjects, although most subjects exhibit greate
st variation in the vertical meridian of the eye on most of the measur
ing occasions. Variance-covariance matrices based on vector h are give
n. In some cases keratometric variation approaches neutral uniform var
iation. In many of the subjects, mean keratometric measurements change
from morning to afternoon, usually showing an increase in curvature l
ater in the day. Physical activity may increase keratometric variation
and mean curvature.