Unsupervised identification of patterns in microarray data has been a productive approach to uncovering relationships between genes and the biological process in which they are involved. Traditional model-based clustering approaches as well as some recently developed model-based mining approaches for integrating genomic and functional genomic data rely on one's ability to determine the correct number of clusters or modules in the data. In this paper we demonstrate that the performance of such methods in general can be significantly improved by accounting for uncertainties inherent to the process of identifying the optimal number of clusters in the data. We demonstrate that the Bayesian averaging approach to clustering via infinite mixture model offers a more robust performance than the traditional finite mixture model in which the optimal number of clusters is determined using the Bayesian Information Criterion. This performance improvement is demonstrated through a simulation st...