[en] The release of fission products from AGR fuel failures and the way in which the signals from such failures must be detected against the background signal from uranium contamination of the fuel is considered. Theoretical assessments of failure detection are used to show the limitations of the existing Electrostatic Wire Precipitator Burst Can Detection system (BCD) and how its operating parameters can be optimised. Two promising alternative methods, the 'split count' technique and the use of iodine measurements, are described. The results of a detailed study of the mechanical and electronic performance of the present BCD trolleys are given. The limited experience of detection and location of two fuel failures in CAGR using conventional and alternative methods is reviewed. The larger failure was detected and located using the conventional BCD equipment with a high confidence level. It is shown that smaller failures may not be easy to detect and locate using the current BCD equipment, and the second smaller failure probably remained in the reactor for about a year before it was discharged. The split count technique used with modified BCD equipment was able to detect the smaller failure after careful inspection of the data. (author)