Turbulence and chemistry interaction (TCI) is still an open problem in the combustion science. Models are needed to account for the unclosed chemistry source terms when the averaged or filtered Navier-Stokes governing equations are solved for chemical species in reactive flows. Finite-rate (FR) combustion models attempting to model the low-pass filtered production/consumption rates (in the case of Large Eddy Simulation or LES) directly. The Eddy Dissipation Concept (EDC) and Scale Similarity (SS) models are two types of FR combustion models for LES. The aim of the study is to assess and improve the performance of the two mentioned models using Direct Numerical Simulation (DNS) databases. In DNS all flow and chemistry scales are resolved and no TCI modeling is required. The DNS databases of reactive flows with relatively detailed chemistry, which are now available thanks to massively large parallel computational tools and codes, can be utilized to both gain fundamental insights on the turbulent reactive flows and directly assess TCI models. The assessment can be through a priori and a posterior DNS analyses. In the current work, fundamental analyses are carried out, using DNS databases of non-premixed jet flames, on the spectral behavior (velocity and dissipation of kinetic energy spectra) and the internal intermittency phenomenon in this type of flames. The physical findings are applied to develop new FR combustion models for LES. In particular, Eddy Dissipation Concept (EDC) is improved by the modifications on the coefficients and the intermittency factor of the model. The modification on the coefficients follows a theoretical basis in which the new findings on spectral behavior have been applied to reduce the degree of freedom of the model by relating the two free coefficients of the model. On the other hand, the modification in the intermittency factor of EDC is the direct application of the observations in the scaling of dissipation fluctuations. The new EDC models are then a priori assessed using the DNS databases. Besides, the existing Scale Similarity (SS) models for LES are a priori assessed using the DNS databases and new dynamic SS models are developed based on Germano's identity for LES and further assessed using the a priori DNS analysis.