A　Fixed　threshold　is　usually　used　for　synchronous　detection　in　traditional　Frequency　Shift　Keying(FSK)　system　based　on　Discrete　Fourier　Transform(DFT).　The　disadvantage　of　the　fixed　threshold　is　that　it　cannot　adjust　itself　dynamically　to　the　change　of　signal　fading　and　noise　jamming　in　a　time-varying　channel,　which　makes　the　system　unsuitable　for　practical　applications.　To　solve　this　problem,　a　novel　adaptive　threshold　demodulation　method　based　on　random　signal　detection　and　estimation　theory　is　proposed.　Then　the　rigorous　theoretical　proofs　are　shown,　and　the　definite　expression　of　the　adaptive　threshold　is　given.　Finally,　simulation　results　show　that　the　adaptive　threshold　can　adapt　to　the　channel　condition　change　under　different　signal-to-noise　ratios(SNR).　And　the　false　alarm　and　detection　probabilities　agree　well　with　the　theoretical　analyses.　The　algorithm　can　effectively　resolve　the　problem　of　the　threshold-choosing　in　the　time-varying　channel.