The　selection　of　representative　ground　motion　intensity　measure　(IM)　and　structural　engineering　demand　parameter　(EDP)　is　the　crucial　prerequisite　for　evaluating　structural　seismic　performance　within　the　performance-based　earthquake　engineering　(PBEE)　framework.　This　study　focuses　on　this　crucial　step　in　developing　the　probabilistic　seismic　demand　model　for　two-story　and　three-span　subway　stations　exposed　to　transverse　seismic　loadings　in　three　different　ground　conditions.　The　equivalent　linearization　approach　is　used　to　simulate　the　shear　modulus　degradation　and　the　increase　in　damping　characteristics　of　the　soil　under　seismic　excitation.　Nonlinear　fiber　beam-column　elements　are　adopted　to　characterize　the　nonlinear　hysteretic　degradation　of　the　subway　station　structure　during　seismic　events.　A　total　of　21　far-field　ground　motions　are　selected　from　the　PEER　strong　ground　motion　database.　Nonlinear　incremental　dynamic　analyses　(IDAs)　are　conducted　to　evaluate　the　seismic　response　of　the　subway　station.　A　suite　of　23　ground　motion　IMs　is　evaluated　using　the　criteria　of　correlation,　efficiency,　practicality,　and　proficiency.　Then,　a　multi-level　fuzzy　evaluation　method　is　employed　to　integrate　these　evaluation　criteria　and　determine　the　optimal　ground　motion　IMs　in　different　ground　conditions.　The　peak　ground　acceleration　and　sustained　maximum　acceleration　are　demonstrated　to　be　the　optimal　ground　motion　IM　candidates　for　shallowly　buried　rectangular　underground　structures　in　site　classes　I,　II,　and　III,　while　the　root-mean-square　displacement　and　compound　displacement　are　found　to　be　not　suitable　for　this　purpose.