Abstract : Concrete is a material widely used for constructions like bridges, nuclear power stations or bunkers. These buildings can be subjected to dynamic loadings such as industrial accidents or projectile-impacts. Consequently a good knowledge of the mechanical behaviour of concrete is a significant safety issue. This work aims to study the damage mechanisms activated in concrete structures under a projectile-impact or a blast loading. First, numerical simulation of impact tests reveals the significance of the dynamic tensile behaviour of concrete targets to simulate accurately their response under impact. Few test data are available in the open literature for strain rates up to 10^2/s. Nevertheless, these results are dispersed: at 100/s, the dynamic increase factor (DIF, dynamic strength to static strength ratio) varies from 5 for one author to 10 for another. The rate sensitivity of concrete has been studied in LPMM over a wide range of strain rates by means of direct tensile tests on a high-speed hydraulic device (10^-5/s <-> 1/s) and using spalling experiments up to 150/s. For this last technique, numerical simulations have been performed to optimize the loading and get a transient but homogeneous tensile loading within the concrete specimen. Moreover computations have been used to evaluate and validate the data processing. Then, experiments have been carried out on two concretes: a microconcrete (MB50) with a fine mesostructure adapted to laboratory testing (= 2mm) and a concrete (R30A7) which is representative of a standard concrete with a compressive strength of 30 MPa and a maximum aggregate size of 8 mm. In each experiment, a particular attention has been devoted to the moisture of the tested specimen. Several spalling experiments have been conducted on dried and re-infiltrated specimens. These experiments have been used to understand the difference of dynamic tensile strength between wet and dry concrete observed by several authors. In parallel these materials have been subjected to edge-on impact tests in two configurations: * Sarcophagus configuration: the concrete plate is encapsulated in an aluminium box allowing keeping fragments near their initial position. The specimen can be recovered post mortem and infiltrated by a hyperfluid resin to reveal the damage induced by the impact. * Open configuration: a high speed camera is used to record the fragmentation process. A digital image correlation (DIC) software called CorreliQ4 developed at LMT Cachan has been applied to realise full-field measurements to identify cracks development and propagation during the test. Another impact test has been proposed: the cratering test has been used to study the penetration of a projectile in a concrete target. Again, the DIC technique allowed performing displacement measurements. It shows that, in this test, the material erosion is a local process. All the experimental data can be used to assess the accuracy of a modelling approach. In this work, the "multiple fragmentation" model proposed by Denoual-Forquin-Hild has been used to simulate spalling and edge-on impact experiments. This model based on a micromechanical description of the fragmentation process allows predicting the maximum tensile strength and the cracking density in the specimen. Each parameter may be evaluated by specific experiments. The model has been implemented in Abaqus/explicit via a user subroutine. Numerical simulations of spalling tests and EOI tests showed a good agreement with experimental results. Nevertheless the modelling did not allow reproducing the cohesion (residual strength) observed in several experiments. To improve the numerical predictions a mesoscopic approach has been employed to simulate the dynamic experiments performed on the standard concrete. In this method, aggregates and matrix are differentiated. Here, aggregates have been supposed perfectly elastic and the behaviour of the matrix phase has been identified using homogenization methods. The presence of heterogeneities at the mesoscale improves the accuracy of numerical predictions. Finally, the influence of the cohesive behaviour of concrete in dynamic conditions has been studied by means of numerical simulations. A micromechanical approach has been used to evaluate the effects of cohesion on the global response of a material subjected to a fragmentation process. Three micromechanical models have been tested. Finally a term has been added to the "multiple fragmentation" model to take into account the cohesion of the damaged concrete. This modification improved the predictions of the dynamic tensile response of concrete.