A direct full configuration interaction approach, previously used for studying individual low‐lying eigenvalues, is combined with iterative Lanczos calculations, in order to obtain global properties of large Hermitean Hamiltonian matrices. To this effect systematic generation of random start vectors is used to compute statistical approximations to the density of states (DOS). Applications for cyclic polyenes in order to illustrate the viability of the scheme modeled by correlated, high‐binding Pariser‐Parr‐Pople (PPP) Hamiltonians of increasing complexity are presented. The degree of correlation in the solutions is controlled by the parameter β. Convergence properties of the DOS for N = 10 and N = 14 are studied varying the extent of correlation. © 1997 John Wiley & Sons, Inc. Int J Quant Chem 63: 719–728, 1997