Progressive supranuclear palsy causes diverse clinical presentations, including classical Richardson's syndrome and several variant phenotypes. Clinical trials of disease-modifying therapies have recently been completed, with more planned for the next 2 years. However, many people with progressive supranuclear palsy do not meet eligibility criteria for these clinical trials. Understanding clinical progression with different phenotypes would improve trial design and enhance the accuracy of risk-benefit and cost-benefit assessments of new treatments for progressive supranuclear palsy. We set out to determine rates of motor and cognitive progression of possible, probable and definite progressive supranuclear palsy, with different phenotypes, from a representative cohort in a regional UK healthcare service. Longitudinal clinical data from people with Richardson's syndrome and variant phenotypes were analysed using linear mixed-modelling, using both the full and modified versions of the Progressive Supranuclear Palsy Rating Scale, Mini-Mental State Examination and the revised Addenbrooke's Cognitive Examination. Subgroup analyses considered patients meeting recent Phase II trial entry criteria and patients with neuropathological confirmation. Two hundred and twenty-seven patients [male = 59%, mean age (±standard deviation), 71.8 (±7.0) years] were followed for a mean 21.6 (±15.6) months. One hundred and seventy-four (77%) had Richardson's syndrome at the outset, 25 had cortical variant presentations (13%, frontal, corticobasal, speech and language variants) and 28 had subcortical variant presentations (14%, parkinsonism, postural instability and gait freezing variants). Across all participants, annual progression in Richardson's syndrome was faster than variant phenotypes on the Mini-Mental State Examination (-1.8 versus -0.9/year, P = 0.005) and revised Addenbrooke's Cognitive Examination (-5.3 versus -3.0/year, P = 0.01) but not the Progressive Supranuclear Palsy Rating Scale (9.0 versus 7.1/year, P = 0.2) nor the modified Progressive Supranuclear Palsy Rating Scale (2.7 versus 2.3/year, P = 0.4). However, for those with more than 1 years' follow-up, a significant difference was observed between Richardson's syndrome and variant phenotypes in Progressive Supranuclear Palsy Rating Scale (8.7 versus 6.3/year, P = 0.04). Survival was longer in variant phenotypes than Richardson's syndrome [7.3 (±3.9) versus 5.6 (±2.0) years, P = 0.02]. Pathologically confirmed cases (n = 49) supported these findings. Patients meeting basic trial-eligibility criteria (n = 129) progressed faster on the Progressive Supranuclear Palsy Rating Scale than trial-not-eligible patients (10.1 versus 6.1/year, P = 0.001). In conclusion, phenotypes other than Richardson's syndrome show slower progression and longer survival. Trial criteria do not select representative progressive supranuclear palsy cases. This has implications for trial design, and application of trial results to clinically more diverse patient populations.