We review the assumptions and the logic underlying the derivation of DLCQ m
atrix models. In particular we try to clarify what remains valid at finite
N, the role of the non-renormalization theorems and higher order terms in t
he supergravity expansion. The relation to Maldacena's conjecture is also d
iscussed. In particular the compactification of the matrix model on T-3 is
compared to the AdS(5)XS(5) N=4 super Yang-Mills duality, and the different
role of the branes in the two cases is discussed. [S0556-2821(99)01904-9].