In a statistical decision problem, the use of approximate or estimated priors leads naturally to the question of the continuity of risk

in the variable

. This continuity condition is explored, and in the finite parameter state case, a characterization in terms of the differentiability of the Bayes envelope is given.