We study asymptotic Bayesian analysis based on a limited information e
stimator, with an unknown partial likelihood function. It is found tha
t the asymptotic distribution of the estimator approximates the poster
ior distribution, provided that the estimator's distribution converges
uniformly in local neigborhoods around the true parameter value. This
provides a Bayesian interpretation to classical limited information p
rocedures, making them available for a semi-parametric Bayesian analys
is. Uniform convergence in distribution is essential for such result,
as illustrated by examples in which the estimators are pointwise asymp
totically normal at every parameter value, but the posterior distribut
ions display discontinuous and possibly non-normal behaviors. (C) 1999
Elsevier Science S.A. All rights reserved.