Introduction: Arthroplasty-associated bone loss remains a clinical problem: stiff metallic implants disrupt load transfer to bone and hence its remodeling stimulus. The aim of this research was to analyze how load transfer to bone is affected by different forms of knee arthroplasty: isolated partial knee arthroplasty (PKA), compartmental arthroplasty (CPKA, two or more PKAs in the same knee) and total knee arthroplasty (TKA). Methods: An experimentally validated subject-specific finite element model was analyzed native, and with medial unicondylar, lateral unicondylar, patellofemoral, biunicondylar, medial bicompartmental, lateral bicompartmental, tricompartmental and total knee arthroplasty. Three load cases were simulated for each: gait, stair ascent and sit-to-stand. Strain shielding and overstraining were calculated from the differences between the native and implanted states. Results: For gait, the TKA femoral component led to mean strain shielding (30%) more than three times higher than PKA (4% to 7%) and CPKA (5% to 8%). Overstraining was predicted in the proximal tibia (TKA 21%, PKA/CPKA 0 to 6%). The variance in the distribution for TKA was an order of magnitude greater than for PKA/CPKA indicating less physiological load transfer. Only the TKA-implanted femur was sensitive to load case: for stair ascent and gait, almost the entire distal femur was strain-shielded, whereas during sit-to-stand the posterior femoral condyles were overstrained. Discussion: TKA requires more bone resection than PKA and CPKA. These finite element analyses suggest that a longer-term benefit for bone is probable as partial and multicompartmental knee procedures lead to more natural load transfer compared to TKA. High-flexion activity following TKA may be protective of posterior condyle bone resorption, which may help explain why bone loss affects some patients more than others. The male and female intact bone models are included as supplementary material.