Joint　inversion　of　horizontal-to-vertical　spectral　ratios　(HVSRs)　and　dispersion　curves　(DCs)　from　seismic　noise　recordings　has　been　extensively　used　to　overcome　the　lack　of　inversion　uniqueness　in　the　noise-based　HVSR　(NHV)　or　DC　inversions　alone.　Earthquake　recordings　contain　information　about　the　structural　properties　of　sedimentary　layers　and　provide　body-wave　data　complementary　to　seismic　noise　recordings　to　estimate　site　velocity　structures,　particularly　in　the　high-frequency　band.　We　propose　a　joint　inversion　of　the　Rayleigh　wave　DC　obtained　from　array　measurements　and　earthquake-based　HVSR　(EHV).　The　EHV　is　derived　from　earthquake　motions　rather　than　from　microtremors　based　on　the　diffuse-field　theory　of　plane　waves.　We　investigated　the　complementarity　of　EHV　and　surface-wave　DC　in　the　joint　inversion　through　sensitivity　analyses.　The　DC　is　sensitive　to　bedrock　shear-wave　velocities　in　the　low-frequency　range　and　is　supplemented　to　some　degree　by　the　EHV　in　the　high-frequency　range.　The　EHV　is　more　sensitive　to　sediment　thicknesses　almost　over　the　entire　frequency　range.　The　joint　inversion　is　implemented　by　a　hybrid　global　optimization　scheme　that　combines　genetic　algorithm　(GA)　and　simulated　annealing　(SA)　to　avoid　premature　convergence　in　the　GA.　The　sensitivity　of　inversion　parameters　was　tested　to　demonstrate　that　the　P-　and　S-wave　velocities　and　thicknesses　of　soil　layers　are　the　dominant　parameters　influencing　EHV　and　DC　responses.　The　proposed　method　was　validated　by　using　synthetic　models　to　compare　the　joint　inversion　with　EHV　or　DC　inversions　alone.　The　joint　inversion　was　applied　to　the　Garner　Valley　Downhole　Array　(GVDA)　data　for　identifying　the　velocity　structures　of　the　site　based　on　earthquake　and　noise　observations.　The　inversion　results　for　the　P-　and　S-wave　velocities　and　thicknesses　of　soil　layers　strongly　suggest　that　the　joint　inversion　is　an　efficient　method　to　estimate　site　velocity　structures.