By associating the Choi matrix form of a completely positive, trace preserving (CPTP) map with a particular space of matrices with orthonormal columns, called a Stiefel manifold, we present a family of parametric probability distributions on the space of CPTP maps that is amenable to Bayesian analysis of process tomography. Using the statistical theory of exponential families, we derive a distribution that has an average Choi matrix as a sufficient statistic, and relate this to data gathered through process tomography. This results in a maximum entropy distribution completely characterized by the average Choi matrix, to our knowledge the first example of a continuous, non-unitary random CPTP map that can capture meaningful prior information about arbitrary errors for use in Bayesian estimation. As specific examples, we show how these distributions can be used for full Bayesian tomography as well as maximum a posteriori (MAP) estimates. These distributions also have relevance in recently proposed importance sampling-based Bayesian procedures for process tomography. As an aside, we show how this Stiefel manifold representation can be used to perform manifold-based optimization that maintains the CP and TP properties along the entire search path without the use of general constraint optimization techniques.