Purpose: To (1) investigate the association between sprint power and aerobic power output (POpeakGXT) during a graded peak exercise test (GXT); and (2) validate the prediction models of POpeakGXT based on sprint power and personal and lesion characteristics. Materials and methods: Wheelchair users with tetraplegia (N = 35) and paraplegia (N = 58) performed a 30 s-Wingate test and GXT on an asynchronous arm-crank ergometer. Data were split into samples to develop and validate the model. Sprint power (POmeanWingate and POpeakWingate, respectively) and POpeakGXT were determined. Regression analyses were performed to develop POpeakGXT prediction models. Candidate independent variables included POmeanWingate or POpeakWingate, age (years), sex, body mass (kg) or BMI (kg/m2), time since injury (TSI, years) and lesion level (tetraplegia/paraplegia). The best model was validated by comparing the predicted POpeakGXT with measured POpeakGXT. Results: The best model (R 2 = 0.76) to predict POpeakGXT included POmeanWingate, BMI and all other independent variables. No significant difference was found between measured (68 ± 35 W) and predicted POpeakGXT (68 ± 30 W, p = 0.97). The ICC was excellent (0.89 with 95% confidence intervals: 0.75–0.95). The 95% limits of agreement for the Bland–Altman plots were wide (−30 to 31 W). Conclusions: Strong associations were found between POmeanWingate and POpeakGXT. Although relative agreement was excellent, absolute agreement was low.Implications for rehabilitation There is a strong relationship between peak aerobic power output and sprint power output, both tested on an arm-crank ergometer, in people with spinal cord injury. A prediction model for peak aerobic power output, based on sprint power output and personal and lesion characteristics, showed a high explained variance. The predictive model can give a guideline for choosing the right graded exercise test protocol but should be used with caution.