We introduce a concept of input-to-output practical stability (IOpS) w
hich is a natural generalization of input-to-state stability proposed
by Sontag. It allows us to establish two important results. The first
one states that the general interconnection of two IOpS systems is aga
in an IOpS system if an appropriate composition of the gain functions
is smaller than the identity function. The second one shows an example
of gain function assignment by feedback. As an illustration of the in
terest of these results, we address the problem of global asymptotic s
tabilization via partial-state feedback for linear systems with nonlin
ear, stable dynamic perturbations and for systems which have a particu
lar disturbed recurrent structure.