Résumé : This work addresses the application of non-premixed filtered tabulated chemistry as a turbulent combustion modeling strategy in the LES framework. On the first part of this study the effects of the filtering operation on non-premixed flamelets are carefully appraised, considering an individual flamelet and the entire manifold. Subsequently, a systematic approach is followed where first the numerical implementation is verified. Afterwards validation is done on a coflow laminar diffusion flame, where promising results encourage the further model appraisal on a more complex turbulent configuration. This is finally achieved under turbulent conditions of Flames D and E, where the formalism including a SGS wrinkling modeling function adequately describes the wrinkled flame front features. The formalism assessment on a laminar coflow diffusion flame reveals a considerable sensitivity to the flame dimensionality. A flame sensor based on the mixture fraction gradient, with a tolerance to take into account the numerical grid resolution, is introduced and proves to deliver satisfactory results. The sensor-determined model activation allows to adequately represent the underlying physics behind flame filtering and so it endorses the consistency of the numerical procedure. The evaluation of the non-premixed FTACLES model on turbulent flames D and E demonstrates that the formalism coupling with a SGS wrinkling modeling function can adequately describe the wrinkled flame front condition. The model performs significantly well employing a three-dimensional tabulation strategy, where the numerical grid is coupled with the model by the third parameter, i.e. the computational cell size. The predictions for both the major stable species and the minor ones accurately correspond with the undergoing physics. The obtained results have a deep theoretical implication for the combustion research. First, they confirm the idea that SGS closure in diffusive combustion can be derived based on filtering arguments, and not only based on statistical approaches. Second, they demonstrate the enormous potential of the non-premixed FTACLES formalism once a sound flame sensor and a SGS wrinkling modeling function are included.