UHF radio frequency identification (RFID) has gathered significant interest in the field of long-distance automatic identification applications. Since UHF RFID shares the frequency band with other RFID and/or other wireless systems, it is important to determine how much interference can be applied without causing a significant degradation of anti-collision speed. In this paper, the permissible link quality for RFID anti-collision in a practical environment is discussed by considering an erroneous communication link, taking into account of bit encoding and the type of interference. We approach the quantification of permissible link quality experimentally along with protocol simulations and the mathematical analyses. An international standard protocol, employing frame slotted ALOHA, was used as the air protocol. For these investigations, the present authors developed a protocol simulator. The simulation results were compared with analytical values based on Poisson distribution. The investigation in the return (tag to reader) link, and the forward (reader to tag) link, were analyzed separately. As result of the protocol simulation, it is generally important to secure the Pulse Error Rate 10^-4 or better in both return and forward links for the anti-collision of 64 or less tags. The quality of the return link may be relaxed when the application does not require fast anti-collision. The degradation of the forward link, on the other hand, may entail loss of important commands, resulting in extremely slow anti-collision. It is measured experimentally that the required link quality can be relaxed by up to 10dB in the return links and by 5dB in the forward link when the primary source of interference originates in the interfering readers.