Abstract Background To determine the validity of a semi-automated segmentation of bone marrow lesions (BMLs) in the knee. Methods Construct validity of the semi-automated BML segmentation method was explored in two studies performed using sagittal intermediate weighted, turbo spine echo, fat-suppressed magnetic resonance imaging sequences obtained from the Osteoarthritis Initiative. The first study (n = 48) evaluated whether tibia BML volume was different across Boston Leeds Osteoarthritis Knee Scores (BLOKS) for tibia BMLs (semiquantitative grades 0 to 3). In the second study (n = 40), we evaluated whether BML volume change was associated with changes in cartilage parameters. The knees in both studies were segmented by one investigator. We performed Wilcoxon signed-rank tests to determine if tibia BML volume was different between adjacent BLOKS BML scores and calculated Spearman correlation coefficients to assess the relationship between 2-year BML volume change and 2-year cartilage morphometry change (significance was p ≤ 0.05). Results BML volume was significantly greater between BLOKS BML score 0 and 1 (z = 2.85, p = 0.004) and BLOKS BML scores 1 and 2 (z = 3.09, p = 0.002). There was no significant difference between BLOKS BML scores 2 and 3 (z = −0.30, p = 0.77). Increased tibia BML volume was significantly related to increased tibia denuded area (Spearman r = 0.42, p = 0.008), decreased tibia cartilage thickness (Spearman r = −0.46, p = 0.004), increased femur denuded area (Spearman r = 0.35, p = 0.03), and possibly decreased femur cartilage thickness (Spearman r = −0.30, p = 0.07) but this last finding was not statistically significant. Conclusion The new, efficient, and reliable semi-automated BML segmentation method provides valid BML volume measurements that increase with greater BLOKS BML scores and confirms previous reports that BML size is associated with longitudinal cartilage loss.