The paper is organized as follows. Section 2 provides elements for Bayesian inference in a finite mixture of log-normal distributions when empirical quantiles are used to elicit prior information and when the marginal likelihood is used to select the optimal number of mixture components. A discussion is led about the influence of this prior on label switching. Section 3 reviews the analytical expressions of commonly used inequality indices, sets out Bayesian inference for these indices and provides the decomposability of the Generalized Entropy index in the framework of mixture models. Section 4 illustrates the approach using the FES data and Section 5 concludes.