Summary Objective Knee replacement (KR) represents a clinically important endpoint of knee osteoarthritis (KOA). Here we examine the 4-year trajectory of femoro-tibial cartilage thickness loss prior to KR vs non-replaced controls. Methods A nested case–control study was performed in Osteoarthritis Initiative (OAI) participants: Cases with KR between 12 and 60 month (M) follow-up were each matched with one control (without KR through 60M) by age, sex, and baseline radiographic stage. Femoro-tibial cartilage thickness was measured quantitatively using magnetic resonance imaging (MRI) at the annual visit prior to KR occurrence (T0), and at 1–4 years prior to T0 (T−1 to T−4). Cartilage loss between cases and controls was compared using paired t-tests and conditional logistic regression. Results One hundred and eighty-nine knees of 164 OAI participants [55% women; age 64 ± 8.7; body mass index (BMI) 29 ± 4.5] had KR and longitudinal cartilage data. Comparison of annualized slopes of change across all time points revealed greater loss in the central medial tibia (primary outcome) in KRs than in controls [94 ± 137 vs 55 ± 104 μm; P = 0.0017 (paired t); odds ratio (OR) 1.36 (95% confidence interval (CI): 1.08–1.70)]. The discrimination was stronger for T−2 → T0 [OR 1.61 (1.33–1.95), n = 127] than for T−1 → T0, and was not statistically significant for intervals prior to T−2 [i.e., T−4 → T−2, OR 0.97 (0.67–1.41), n = 60]. Results were similar for total medial femoro-tibial cartilage loss (secondary outcome), and when adjusting for pain and BMI. Conclusions In knees with subsequent replacement, cartilage loss accelerates in the 2 years, and particularly in the year prior to surgery, compared with controls. Whether slowing this cartilage loss can delay KR remains to be determined.