我們參考 Schmidt and Sickles （ 1984 ）及 Cornwell， et al.（ 1990 ） 所提出分析 Panel 資料的隨機邊界生產函數估計方法， 於文中以隨機邊界成本函數來進行 效率分析，並計算出各中小企銀每年的相對效率值。接著於第二階考慮銀行業的特性變數， 以隨機效果模型探討影響銀行效率的原因。 本文研究有下列幾項重要結果： 第一， 在 1986-94 年間，臺南企銀的平均成本效率水準最高，達到 78.65%， 而新竹企銀的表現最差 為 73.11%，然而，不同企銀彼此之間效率差異並不太大。 第二，由隨機成本函數估計結果 可知，影響效指標的時間、時間平方項估計係數顯著地異於零。這充份顯示銀行的效率水準 在分析期間並非一成不變。第三，催收款比例和股權分散程度愈高對中小企銀之成本率水準 有不良影響，而自有資金比例及分行數愈多則有助於成本效率水準的提升。

　　　　　In this paper we apply a panel data model from Cornwell, Schmidt, and S ickles (1990) to analyze the cost efficiency of six SME banks in Taiwan for 1986-1994. Our model allows us to estimate time-varying efficiency levels for an individual bank without imposing strong distributional assumptions on the error terms. The relative efficiencies among these banks are calculated each year and the determinants of the bankinf efficiency are found by the regression of the random-effect model. Our major findings are:(1) On average, the level of the cost efficiency for Tainan SME bank was the highest, while that for the Hsinchu SME bank was the lowest. The difference in the efficiency level among these six SME banks was not very large.(2) The estimated coefficients of the time and time-squared terms in the stochastic frontier cost function show that they are significantly different from zero, which implies that the efficiency level of the SME banks varied over time. (3)The higher the ratio of nonaccrual loans to total loans and the deconcentration ratio of stock ownership are, the lower the efficiency of the SME banks were. On the other hand, the increases in the percentage of the financial sources from stockholders, and the number of branches played important roles in promoting the cost efficiency of the SME banks.