A matrix variate Dirichlet vector is a random vector of independent Wishart matrices 'divided' by their sum. Many properties of Dirichlet vectors are usually established through integral computations assuming existence of density for the Wishart's. We propose a method to deal with the general case where densities need not exist. On the other hand, in dimension larger than 2, Dirichlet processes reduce to Dirichlet vectors and posterior of Dirichlet need not be Dirichlet.