Failure location in transparent optical networks is difficult because of the large amount of alarms that a failure can trigger and because of corrupted alarms. One problem that network operators often face is how to set the thresholds in monitoring devices. Setting the thresholds low results in false alarms, whereas setting them high presents the risk of missing a significant degradation in network performance because of missing alarms. In this work, we study the time complexity of the failure location problem in transparent optical networks and provide an efficient algorithm to solve this problem. More significantly, we show that for a network with binary alarms (alarms are either present or not), there is an asymmetry between false and missing alarms. We prove that false alarms can be corrected in polynomial time, but that the correction of missing alarms is NP-hard. Because of this asymmetry between false and missing alarms, false alarms have lesser effect on the accuracy of the diagnosis results than missing alarms do. Network operators therefore, when allowed, should set the threshold low to favor false alarms rather than high.