Time-domain reflectometry (TDR), the most widely used testing method for fault-detection in cables, is tested against soft faults of increasing severity. The intensity of TDR echoes is proven to be an unreliable estimator of fault severity, since both the bandwidth of testing signals and the length of the fault have a strong impact on the results. Moreover, it is proven that faults of very different severity may generate virtually identical echoes making it impossible to assess how critical a fault is. Simple frequency-domain estimators are introduced, based on models of a fault reflectivity, leading to minimum requirements for the test bandwidth in order to accurately identify soft faults. As a practical consequence, it is concluded that only faults above a critical length can be accurately identified, implying that no clear decision can be taken about shorter faults.