Introduction There are conflicting data on the natural history and clinical significance of bone marrow lesions (BMLs). The aims of this study were to describe the natural history of MRI-detected BMLs at the knee using a quantitative measure and examine the association of BMLs with pain, function and stiffness scores, and total knee replacement (TKR) surgery. Methods A total of 395 older males and females were randomly selected from the general population (mean age 63 years, range 52 to 79) and measured at baseline and approximately 2.7 years later. BMLs were determined using T2-weighted fat saturation MRI by measuring the maximum area of the lesion. Reproducibility was excellent (intraclass correlation coefficient (ICC): 0.97). Pain, function, and stiffness were assessed by Western Ontario and McMaster Universities Osteoarthritis (WOMAC) scores. X-ray was used to assess radiographic osteoarthritis (ROA) at baseline. Results At baseline, 43% (n = 168/395) had a BML. Of these 25% decreased in size and 24% increased. Of the remaining sample (n = 227), 7% developed a new BML. In a multivariable model, a change in BML size was associated with a change in pain and function scores (β = 1.13 to 2.55 per 1 SD increase, all P < 0.05), only in those participants without ROA. Lastly, baseline BML severity predicted TKR surgery (odds ratio (OR) 2.10/unit, P = 0.019). Conclusions In a population based sample, BMLs (assessed by measuring maximal area) were not static, with similar proportions both worsening and improving. A change in BML size was associated with changes in pain in those without established ROA. This finding suggests that fluctuating knee pain may be attributable to BMLs in those participants with early stage disease. Baseline BMLs also predicted TKR surgery. These findings suggest therapeutic interventions aimed at altering the natural history of BMLs should be considered.