The widespread use of the Mixed Multinomial Logit model, in the context of discrete choice data, has made the issue of choosing a mixing distribution very important. The choice of a specific distribution may seriously bias results if that distribution is not suitable for the data. We propose a flexible hierarchical Bayesian approach in which the mixing distribution is approximated through a mixture of normal distributions. Numerical results on a real data set are provided to demonstrate the usefulness of the proposed method.