Selective expression of a gene product (mRNA or protein) is a pattern in which the expression is markedly high, or markedly low, in one particular tissue compared with its level in other tissues or sources. We present a computational method for the identification of such patterns. The method combines assessments of the reliability of expression quantitation with a statistical test of expression distribution patterns. The method is applicable to small studies or to data mining of abundance data from expression databases, whether mRNA or protein. Though the method was developed originally for gene-expression analyses, the computational method is, in fact, rather general. It is well suited for the identification of exceptional values in many sorts of intensity data, even noisy data, for which assessments of confidences in the sources of the intensities are available. Moreover, the method is indifferent as to whether the intensities are experimentally or computationally derived. We show details of the general method and examples of computational results on gene abundance data.