Monte Carlo criticality calculation allows to estimate the effective mutiplication factor as well as local quantities such as local reaction rates. Some configurations presenting weak neutronic coupling (high burn up profile, complete reactor core, ...) may induce biased estimations for keff or reaction rates. In order to improve robustness of the iterative Monte Carlo méthods, a coupling with a deterministic code was studied. An adjoint flux is obtained by a deterministic calculation and then used in the Monte Carlo. The initial guess is then automated, the sampling of fission sites is modified and the random walk of neutrons is modified using splitting and russian roulette strategies. An automated convergence detection method has been developped. It locates and suppresses the transient due to the initialization in an output series, applied here to keff and Shannon entropy. It relies on modeling stationary series by an order 1 auto regressive process and applying statistical tests based on a Student Bridge statistics. This method can easily be extended to every output of an iterative Monte Carlo. Methods developed in this thesis are tested on different test cases.