For the Current Employment Statistics Program, approximately unbiased and stable variance estimators are important for the empirical evaluation of standard design-based point estimators, and for production of related small domain estimators. In some cases, standard design-based variance estimators can be relatively unstable, which may lead to consideration of alternative variance estimators based on generalized variance functions. This paper presents an exploratory analysis of generalized variance function models for estimates of total monthly employment with domains determined by the intersection of metropolitan statistical area and major industrial division. Three topics receive principal attention: a.) a detailed description of features of the underlying sample design that are important in variance estimation; b.) graphical evaluation of potential biases in generalized variance function estimators; and c.) omnibus measures of the relative magnitudes of the fixed and random components of model lack of fit.