Purpose: To develop and validate predictive models for peak power output to provide guidelines for individualized handcycling graded exercise test protocols for people with spinal cord injury (SCI); and to define reference values.Materials and methods: Power output was measured in 128 handcyclists with SCI during a synchronous handcycling exercise test. Eighty percent of the data was used to develop four linear regression models: two theoretical and two statistical models with peak power output (in W and W/kg) as dependent variable. The other 20% of the data was used to determine agreement between predicted versus measured power output. Reference values were based on percentiles for the whole group.Results: Lesion level, handcycling training hours and sex or body mass index were significant determinants of peak power output. Theoretical models (R2 = 42%) were superior to statistical models (R2=39% for power output in W, R2 = 30% for power output in W/kg). The intraclass correlation coefficients varied between 0.35 and 0.60, depending on the model. Absolute agreement was low.Conclusions: Both models and reference values provide insight in physical capacity of people with SCI in handcycling. However, due to the large part of unexplained variance and low absolute agreement, they should be used with caution. Implications for rehabilitationIndividualization of the graded exercise test protocol is very important to attain the true peak physical capacity in individuals with spinal cord injury.The main determinants to predict peak power output during a handcycling graded exercise test for individuals with a spinal cord injury are lesion level, handcycling training hours and sex or body mass index.The predictive models for peak power output should be used with caution and should not replace a graded exercise test.