This dissertation takes place in the framework of design and analysis of computer experiments. More precisely, its main focus is on optimization strategies based on surrogate models of the objective function, or metamodels. Its principal motivation is to expose and strengthen existing works on Kriging-based optimization. Some relationships between different classical metamodels are adressed, and some light is shed on the versatility of Kriging and its suitability for sequential and parallel optimization. After a detailed introduction to Kriging (end of part I), several tracks for the enrichment of this metamodel are proposed in part II. Part III is dedicated to some novelties in Krigingbased optimization, in particular concerning the integration of a mixture of metamodels or the parallelisation of evaluations for synchronous distributed computing.
