There are risks in Wireless Local Area Network(WLAN) because of the open channel environment and the traditional key authentication mechanism. Radio frequency fingerprinting identification which extracts hardware features of wireless devices for authentication, could greatly improve the wireless network security. Based on Universal Software Radio Peripheral(USRP) and GNU Radio open source platform, carrier frequency offset of IEEE 802.11a/g signals is extracted as the fingerprint, and the neural network classifier is used for recognition. Firstly, this method collects IEEE 802.11 a/g signals and extracts the carrier frequency offset of each frame, then trains a neural network classifier. Lastly it identifies wireless devices by using the classifier. In two typical indoor environments of the office and the gymnasium, the recognition rate of wireless devices is more than 90%. The experimental results show that wireless devices can be identified by extracting carrier frequency offset of signals based on software radio, and illegal device access can be detected, which could improve the security of wireless network.