As an important province of the less developed regions, the urbanization process of Gansu Province continues to accelerate, the urbanization rate continuously improves, however, the urban efficiency did not attract enough attention. It’s significant to improve the urban competitiveness and sustainable development of less developed regions by increasing the urban efficiency. The data envelopment analysis (DEA) is the effective tool to measure urban efficiency, but the traditional DEA can not distinguish the difference among the effective units and define the urban efficiency growth potential. The aggressive cross-evaluation mechanism is introduced along with virtual decision making units (DMU) to overcome the shortcoming of the traditional DEA. In this paper, with land, capital, labor, technology, information, water and electricity consumption being the input indicators, and GDP the output indicator, the DEA-Cross Model is employed to analyze the urban efficiency of 12 prefecture-level cities in Gansu Province based on the panel dataset of urban socioeconomy from 2005 to 2009. The result of the traditional DEA showed that six cities of Gansu Province maintain DEA effective during the study period, such as Lanzhou, Jinchang, Wuwei, Pingliang, Qingyang and Longnan, accounting for 50% of the total cities. The results of the DEA-Cross Model showed that the urban efficiency in Gansu Province is low (between 0.053 and 0.067) from 2005 to 2009, less than 7% of the ideal DMU. And the standard differential obtained an upward trend from 2005 (0.0476) to 2009 (0.0494), the gap among cities became widened. From the perspective of spatial distribution, cities in the central part of Gansu and in Hexi Region have higher urban efficiency than those in the east and south of Gansu, the gap between different regions became narrow (the standard differential decreased from 0.024 in 2005 to 0.021 in 2009). From the perspective of city type, urban efficiency of industrial cities is higher than that of non-mining cities. From the perspective of city scale, urban efficiency of big cities is higher than that of small and medium-sized cities, and the gap between different scales also became narrow (from 0.051 in 2005 to 0.016 in 2009). The clustering results showed that Lanzhou, Jinchang and Jiayuguan belong to the "high input and high output"type, Dingxi and Longnan belong to the "high input and low output"type, and the remaining cities belong to the" low input and low output" type. On the basis of the present study, some suggestions regarding improving urban efficiency were given in the current situation of urban construction and management, including strengthening land intensive productivity capacity, optimizing the industrial structure, promoting the rational division between different regions, accelerating the process of infrastructure construction, innovative management systems and mechanisms, application of the advanced technologies and optimizing the development environment.