Pain is the main impetus for osteoarthritis (OA) patients to seek healthcare including joint replacement. The pain experience in OA is heterogeneous and affected by factors across multiple domains—peripheral, psychological, and neurological. This indicates the existence of homogenous subgroups/phenotypes within OA patients with pain. We recently identified three pain phenotypes using a wide spectrum of pain-related factors, including structural damage on magnetic resonance imaging (MRI), emotional problems, number of painful sites, sex, body mass index, education level and comorbidities (i.e., Class 1: high prevalence of emotional problems and low prevalence of structural damage (25%); Class 2: low prevalence of emotional problems and high prevalence of structural damage (20%); Class 3: low prevalence of emotional problems and low prevalence of structural damage (55%)). This study was to examine whether the total knee replacement (TKR) risk over 12 years was different among these three pain phenotypes. Data on 963 participants (mean age 62.8 ± 7.4 years) from a population-based cohort study were utilised. Data on socio-demographic, psychological and comorbidities were collected. MRI of the right knee structural pathology was performed. TKR history was ascertained by linking to the Australian Orthopedic Association National Joint Replacement Registry. Latent class analysis and the Cox proportional hazards model were applied for the analysis. During the follow-up period, 41 right and 44 left TKRs in 67 participants were identified. In multivariable analyses, participants in Class 1 and 2 had a higher risk of having a TKR (Class 1 vs. Class 3, HR (hazard ratio) 4.81, 95%CI (confidence interval) 2.33–9.93; Class 2 vs. Class 3, HR 9.23, 95%CI 4.66–18.30). These associations were stronger in the imaged right knee but were also significant in the left knee. Participants within distinct pain phenotypes have different risks of TKR, suggesting that the identified phenotypes reflect distinct clinical subgroups with different prognoses. The risk for TKR was higher in Class 1 than that in Class 3, suggesting that pain/emotional status is a stronger driver for TKR than structural damage, and that selecting patients for TKR needs to be optimized in clinical practice.