The complex impregnation of a multifilament yarn by a cementitious matrix leads to a difficult prediction of the mechanical behaviour of textile reinforced concrete and its less spread use than steel-reinforced concrete. To solve this prob- lem, several models were established but they are not based on direct observa- tions of the embedded yarn or quantification of its impregnation. In order to improve those models, a double resin impregnation process followed by confo- cal microscopy was set up, after pullout test was performed on each sample. Sev- eral parameters were then computed from the obtained images, which enables to quantify the impregnation ofthe yarn for each sample and to compare it with the pullout mechanical results. The number of fully impregnated filaments is found to be the crucial point to explain the pullout maximum load. The type of fail- ure is also defined using those same parameters, computed along the embedded length, and it was found that the shape of the extracted volume of yarn is cylin- drical and so the failure of the filaments is not telescopic. A pre-existing model was then improved, considering all those conclusions obtained by microscopic observations, and a good match between the numerical and the experimental results was found. Multifilament yarns are a continuous and flexible textile structure composed of hundreds of filaments that are maintain together using a chemical product called sizing. Those multifilament yarns can be used to reinforce mortar or con- crete, like steel in steel-reinforced concrete. The performances of those types of composites depend strongly on the strength of the reinforcement/mortar bond. In case of textile /mortar composite, this bond depends on the level of penetra- tion of the mortar into the yarn, so the impregnation of the yarn by the mortar. Since they are both very heterogeneous material, this impregnation is random and incomplete. Consequently, it is very difficult to predict the strength of this bond and so, the strength of the yarn/mortar composite. As a result, this type of composite cannot be widely used. Some models were established to predict the strength ofthose composite; however, they are not based on direct observation of the textile/mortar bond. To improve those models, a new visualisation method of 2 SLAMA et al. the impregnation ofthe yarn by the mortar was set up. A double moulding ofthe impregnated yarn by the mortar was manufactured, using resins and two differ- ent fluorescent dyes, after underwent a mechanical test of pulling out the yarn from the mortar (pullout test). First, a resin with a red fluorescent dye was used for the moulding ofthe yarn shortly after the test, and, second, after dissolution of the mortar around the yarn in acid, a resin with a green fluorescent dye. A num- ber of cross sections of the yarn is observed using a microscope that detects those two dyes and the obtained images are analysed allowing a clear differentiation, location and counting of fully impregnated filaments of the yarn by the mortar and those partially or not impregnated. The results are compared to the mechan- ical parameters of the pullout test. The number of fully impregnated filaments is found to be crucial to explain the strength of the composite. The behaviour dur- ing the pullout test is also explained using parameters computed along the length of the yarn in the mortar. It is found that filaments extracted by pulling-out take a cylindrical shape and so the pullout behaviour is not conical shape as it was suggested in a number of models. Considering all those conclusions obtained by observations, a pre-existing model is improved, and a good match between the numerical and experimental results is found.