Abstract: Numerous unreliable, redundant and time-stamp disordered data take place in RFID applications. However, the existing techniques of RFID data cleaning are mainly for the single-type errors and not adaptable to the complicated errors. An integrated data cleaning mechanism is therefore proposed to provide the accuracy, conciseness and sequentiality for RFID data. The mechanism is composed of local filter and global filter. The local one is used to process the data received by single reader and sequence them in accordance to time-stamp with time delayed, then the false-positives are deleted through corresponding constraints according to different distributions of RFID data stream. On the other hand, the global one is used to process the data from multiple readers, where both the false negatives or positives are filled in or deleted, respectively, considering spatio-temporal correlation of tag readings with constraints set up to cancel redundant data. Experimental results showed that the cleaning mechanism proposed can correct various dirty data efficiently in accordance to different stream distributions.