We present a diagnosis system which combines the outputs of several classifiers performing local generation on the French Telephone Network. This data fusion process takes into account alarms occurring at different times and space locations. This allows to considerably improve upon the performances of a previous diagnosis system which relies only on local decisions.