Abstract : Reactivity measurements by the oscillation technique, as those performed in the Minerve reactor, enable to access various neutronic parameters on materials, fuels or specific isotopes. Usually, expected reactivity effects are small, about ten pcm at maximum. Then, the modeling of these experiments should be very precise, to obtain reliable feedback on the pointed parameters. Especially, calculation biases should be precisely identified, quantified and reduced to get precise information on nuclear data. The goal of this thesis is to develop a reference calculation scheme, with well quantified uncertainties, for in-pile oscillation experiments. In this work are presented several small reactivity calculation methods, based on deterministic and/or stochastic calculation codes. Those method are compared thanks to a numerical benchmark, against a reference calculation. Three applications of these methods are presented here: a purely deterministic calculation with exact perturbation theory formalism is used for the experimental validation of fission product cross sections, in the frame of reactivity loss studies for irradiated fuel; an hybrid method, based on a stochastic calculation and the exact perturbation theory is used for the readjustment of nuclear data, here 241Am; and a third method, based on a perturbative Monte Carlo calculation, is used in a conception study.