Abnormal　joint　loads　in　vivo　accelerate　the　development　of　knee　osteoarthritis　and　lead　to　functional　degradation　of　the　biological/artificial　knee.　Quantification　of　double　condyles　contact　force　of　knee　joint　is　crucial　for　clinical　gait　assessment.　In　this　paper,　muscle　moment　was　calculated　based　on　Hill　model　and　external　knee　adduction　moment　(KAM)　was　estimated　by　using　back　propagation　neural　network　(BPNN).　Considering　both　contributions,　the　knee　contact　force　(KCF)　was　calculated.　The　whole　body　musculoskeletal　model　was　established　based　on　OpenSim,　with　three　degrees　of　freedom　(DOF)　and　double　contact　points　at　knee　joint.　The　reliability　of　the　model　was　verified　by　21　gait　trials　at　free　pace　(1.19±0.09m/s)　of　four　subjects　who　underwent　total　knee　arthroplasty　(TKA).　The　results　showed　that　the　value　of　proposed　modeling　estimation　was　highly　correlated　with　those　of　implant　measurement　in　the　surgical　knee.　The　Pearson　Correlation　Coefficients　(ρ)　of　the　medial/lateral　knee　contact　force　(MKCF/LKCF)　were　0.93±0.06　and　0.81±0.16.　Peak　error　and　time　error　were　less　than　0.40　body　weight　(BW)　and　50ms.　The　estimation　of　proposed　modeling　was　also　highly　correlated　with　simulation　of　OpenSim　in　the　healthy　knee.　It　was　concluded　that　reliable　KCF　could　be　obtained　using　the　proposed　model.　©　2022　IEEE.