Knee osteoarthritis (OA) is a heterogeneous disease, and identification of its subgroups/phenotypes can improve patient treatment and drug development. We aimed to identify homogeneous OA subgroups/phenotypes using pain development over time; to understand the interplay between pain and functional limitation in time course, and to investigate subgroups' responses to available pharmacological and surgical treatments. We used group-based trajectory modelling to identify pain trajectories in the phase-three VIDEO trial (n=474, three-year follow-up) and also in the Osteoarthritis Initiative cohort study (n=4796, nine-year follow-up). We extended trajectory models by (1) fitting dual trajectories to investigate the interplay between pain and functional limitation over time, and (2) including analgesic use as a time-varying covariate. Also, we investigated the relationship between trajectory-groups and knee replacement in regression models. We identified four pain trajectory-groups in the trial and six in the cohort. These overlapped and led us to define four OA phenotypes: low-fluctuating, mild-increasing, moderate-treatment-sensitive and severe-treatment-insensitive pain. Over time, functional knee limitation followed the same trajectory as pain with almost complete concordance (94.3%) between pain and functional limitation trajectory-groups. Notably, we identified a phenotype with severe pain that did not benefit from available treatments, and another one most likely to benefit from knee replacement. Thus, knee OA subgroups/phenotypes can be identified based on patients' pain experiences in studies with long and regular follow-up. We provided a robust approach, reproducible between different study designs that informs clinicians about symptom development and delivery of treatment options and opens a new avenue toward personalized medicine in OA.