Models of disorder with a direction (constant imaginary vector potenti
al) are considered. These non-Hermitian models can appear as a result
of computation for models of statistical physics using a transfer-matr
ix technique, or they can describe nonequilibrium processes. Eigenener
gies of non-Hermitian Hamiltonians are not necessarily real, and a joi
nt probability density function of complex eigenvalues can characteriz
e basic properties of the systems. This function is studied using the
supersymmetry technique, and a supermatrix sigma model is derived. The
sigma model differs from that already known by a new term. The zero-d
imensional version of the sigma model turns out to be the same as the
one obtained recently,for ensembles of random weakly non-Hermitian or
asymmetric real matrices. Using a new parametrization for the supermat
rix Q, the density of complex eigenvalues is calculated in zero dimens
ion for both the unitary and orthogonal ensembles. The function is dra
stically different in these two cases. It is everywhere smooth for the
unitary ensemble but has a delta-functional contribution for the orth
ogonal one. This anomalous part means that a finite portion of eigenva
lues remains real at any degree of the non-Hermiticity. All details of
the calculations are presented.