We study some properties of a class of real-valued, continuous-time random processes, namely multifractal random walks. A striking feature of these processes lie in their scaling property : the distribution of the process at small scale is the same as the distribution at large scale, given some random multiplicative factor independent of the process. The first part of the dissertation deals with the convergence of the empirical moment of the increment of the process in a rather general asymptotic setting where the step of the increment may go to zero while the observation horizon may also go to infinity. In the second part, we propose a family of nonparametric tests that separate multifractal random walks from Itô semi-martingales. After showing the consistency of these tests, we study their behavior on simulations.In the third part, we build a skewed multifractal random walk process, such that the past increment is negatively correlated with the future squared increment. Such a "leverage effect" is notably seen on financial stock and index prices. We compare the empirical properties of this process with real data. The fourth part deals with the parametric estimation of the process. We first show that under certain conditions, one can not estimate two of the three parameters, even if the sample path is continuously observed on some interval. We next study the theoretical and empirical performances of some estimators of the third parameter, the intermittency coefficient, in a Gaussian case