If G is a connected algebraic
linear group with unipotent radical U, Borel and Tits define a Levi factor
of G to be any connected reductive subgroup L of G such that G = L.U
(semidirect product in the sense of algebraic groups). This differs from the
usual notion of Levi decomposition in Lie theory but leads to equivalent
results at characteristic 0. The existence of Levi factors at characteristic
p is problematic, in view of an example of a group having no Levi factor
constructed by Chevalley (unpublished). In this note sufficient conditions are
given for a Levi factor to exist, based on the structure of the Lie algebra of
G.