Data Envelopment Analysis (DEA) is one technique that is commonly used for measuring efficiency. Previous studies focus on identifying the sources of inefficiency with implicit assumption of homogeneity between efficient and inefficient farms. Since technical efficiency cannot be greater than one in the standard DEA, those studies are not capturing the marginal effects of input variables on the efficiency improvement for efficient farms, which leaves the heterogeneity in the marginal effects hard to be examined. We exploit a methodology called super DEA, which facilitates the study on the heterogeneity by producing the ranking information in the efficient Decision Making Units(DMUs). A quantile regression is employed to identify the sources of efficiency for efficient farms and inefficient farms, respectively. Using the Kansas Farm Management Association (KFMA), we find that an increase in feed affects efficiency in opposite ways for efficient farms and inefficient farms. This result suggests that efficient and inefficient farms are heterogeneous.