We present a modified finite temperature Lanczos method for the evaluation of dynamical and static quantities of strongly correlated electron systems that complements the finite temperature method (FTLM) introduced by Jaklič and Prelovšek for low temperatures. Together they allow accurate calculations at any temperature with moderate effort. As an example we calculate the static spin correlation function and the regular part of the optical conductivity σ reg (ω) of the one dimensional Hubbard model at half-filling and show in detail the connection between the ground state and finite temperature method. By using Cluster Perturbation Theory (CPT), the finite temperature spectral function is extended to the infinite system, clearly exhibiting the effects of spin-charge separation. [14,15]. In principle this method can be applied at all temperatures, but at low temperatures the required number of random samples is very large. FTLM is restricted to small systems and particularly at low temperatures, finite size effects become important. They can be overcome, at least for properties derived from the single particle Green's function, by using Cluster Perturbation Theory (CPT) [16,17], which provides infinite system results with remarkable accuracy [18]. However, CPT requires finite cluster Greens functions G ab (ω) for all sites a, b, increasing the number of required Lanczos runs and matrix elements by a factor equal to the number of lattice sites. It therefore requires a sufficiently fast low temperature method.In this paper we present a modified finite temperature Lanczos method which allows to calculate properties for large Hilbert spaces at low temperatures that are not accessible by the existing method. We show that a combination of our low temperature Lanczos method (LTLM) with the FTLM allows an accurate calculation of thermodynamic properties at any temperature with moderate effort.Let us first present the existing FTLM. For the case of a static expectation value of an operator Owith β = 1/T (k B =h = 1) and a sum over a complete orthonormal basis set |n , the FTLM approximation iswith a sum over symmetry sectors s of dimension N s , and R random vectors |r in each sector [19]. M is the number of Lanczos steps. For each random vector |r , a Lanczos procedure is performed, yielding M eigenenergies ε For dynamical correlation functions C(t) = A(t)B , FTLM calculatesHere, an initial vectoris used to generate additional eigenenergiesε and similarly for (3). Thus the ground state result will suffer from severe statistical fluctuations, although the exact (Lanczos) eigenvector |Ψ 0 is reached with every |r and one random vector should be sufficient. Yet, FTLM gets worse with decreasing temperature T .