We propose a decoy-state method to overcome the photon-number-splitting attack for Bennett-Brassard 1984 quantum key distribution protocol in the presence of high loss: A legitimate user intentionally and randomly replaces signal pulses by multi-photon pulses (decoy-states). Then they check the loss of the decoy-states. If the loss of the decoy-states is abnormally less than that of signal pulses, the whole protocol is aborted. Otherwise, to continue the protocol, they estimate loss of signal multi-photon pulses based on that of decoy-states. This estimation can be done with an assumption that the two losses have similar values, that we justify.