The four dimensional variational data assimilation method and the ensemble Kalman filter technique are two promising algorithms for the assimilation of atmospheric observations. The performance of the two algorithms are compared in the context of radar data assimilation at the convective scale. The two methods are tested with a storm simulated by a cloud model initialized with a warm bubble. The 4D-Var method used here is based on a cost function that measures the model fit to the observations over an assimilation window. The model is used as a strong constraint and the adjoint equations are used in the gradient descent method during the minimization of the cost function. The ensemble method derives from the formulation of the ensemble square-root filter, which does not require perturbed observations. Assuming that the observational errors are uncorrelated, the observations are assimilated one at a time by the filter, rendering the analysis step simple. The sensitivity of both methods to different kind of model and observational errors is investigated in terms of analysis and forecast accuracy.