1. A new look at the Bayes procedure
2. Defines a prior to be impartial if it is uniform in a homogeneous parameterization. A locally homogeneous parameterization can be found and this leads to (1)
3. Suggests that improper priors be regarded as limits of data dependent proper priors. Considers an example of a strong inconsistency (sec. 4.2.1) and an example of a marginalization paradox (sec. 4.2.4) and in each case argues that the paradoxes are best resolved by using a sequence of proper priors that depends on the data