This　paper　investigates　the　optimal　intensity　measure　(IM)　for　underground　structure　in　probabilistic　seismic　demand　models.　A　two-dimensional　finite　element　model　that　accounts　for　the　soil-structure　interaction　for　a　subway　station　was　established　employing　the　program　ABAQUS.　Probabilistic　seismic　demand　analysis　was　conducted　based　on　the　cloud　method,　which　involved　248　non-linear　dynamical　analyses.　The　peaks　of　first　and　second　inter-story　drift　ratios,　peaks　of　slab　acceleration,　and　the　peaks　of　bending　moments　for　key　sections　were　recorded　as　the　engineering　demand　parameters　(EDP).　Based　on　the　conventional　and　updated　metrics　of　efficiency,　practically,　proficiency,　16　candidate　intensity　measures　were　evaluated,　and　their　sufficiency　was　verified　through　the　p-value　or　relative　sufficiency　measure.　The　results　demonstrate　that　the　selection　of　EDPs　has　a　minor　influence　on　the　study　of　optimal　IMs;　there　were　no　significant　differences　in　selecting　the　optimal　IMs　based　on　evaluation　considering　either　conventional　or　updated　metrics.　The　sufficiency　test　results　indicated　that　the　assessment　of　p-value　method　is　opposite　to　that　of　the　relative　sufficiency　measure　method　for　some　IMs,　and　that　the　sufficiency　measure　method　could　provide　more　IM　sufficiency　information.　The　main　findings　are　that　the　IMs　related　to　acceleration　curves　performed　well　in　all　metrics　evaluation,　and　that　the　PGA　is　the　most　optimal　IM　for　the　probabilistic　seismic　demand　models　of　underground　frame　structure.