The Bureau of Labor Statistics (BLS) publishes Current Employment Statistics Program employment estimates for many small domains, such as an industry classification within a state or metropolitan statistical area (MSA). To give users a sense of the quality of these estimates, it is important to publish estimates of their associated standard errors. The BLS uses a balanced repeated replication (BRR) technique to produce variance estimates for national industry classification cells and higher levels of aggregation. However, in our limited simulation study, the BRR variance estimator has been found to be unstable, even at the statewide industrial supersector level. In this paper, we propose an empirical linear Bayes (ELB) method for small areas to combine the design-based variance estimator with a suitable model-based estimator. Our preliminary findings are encouraging and motivate further research in this area.