In one embodiment, the present invention is a method for detecting an echo path change (EPC) in a telecommunications network. The method detects whether the effectiveness of echo cancellation of an echo canceller has decreased relatively significantly. Once a relatively significant decrease is detected, the method determines whether the decrease was an EPC or an inadvertent detection of double talk (DT). In particular, the method considers whether echo is effectively cancelled over a hangover period. Further, echo return loss (ERL) estimates are generated over the hangover period and compared to a lowest-possible ERL for the network. If both (1) echo cancellation is ineffective and (2) a sufficient number of ERL estimates are not below the worst-case ERL, then an EPC decision is made. If either (1) echo cancellation is effective or (2) a sufficient number of ERL estimates are below the worst-case ERL, then a DT decision is made.