The paper describes an empirical study O/software maintenance that was carried out in a system software department in 1989 and 1990. The study Jocused on error occurrence andJault detection. Over 400 problem reports were studied. The study showed some unexpected results. It showed, for example, no relation between the phase of error occurrence and the solution time. An explanation is the gap between the methods as they are supposed to be applied and reality. Assessment O/the size o[the gap is one O/'the contributions ~[ this kind Of empirical stu~tv.