A matrix inequality is established that provides an upper bound for a
quadratic form that involves the difference between two linear unbiase
d estimators of the same linear parametric function in a general linea
r model, Various special cases of the inequality are discussed. Certai
n inequalities that arise in the problem of outlier detection and pred
iction of observations come out as special cases, In addition, some ex
tensions of Samuelson's inequality are also obtained.