This paper examines the efficiency of electric power generation plants in the United States. A 1996 data set from the Utility Data Institute and county-level wage data from the Bureau of Labor statistics provide the information needed to construct measures of cost, output, and input prices for 78 steam plants using natural gas as the primary fuel. This paper uses a Bayesian stochastic frontier model that imposes concavity and monotonicity restrictions implied by microeconomic theory to measure efficiency, price elasticities, and returns to scale of these plants. Results indicate that plants on average could reduce costs by up to 13% by eliminating production inefficiency. Results also indicate that most plants operate at increasing returns to scale, suggesting further cost savings could be achieved through increasing output.