Purpose: We estimated the agreement of a thigh-worn accelerometer, the activPAL, used to measure activity and sedentary parameters, with observed mobility assessments of intensive care unit (ICU) survivors. Method: We prospectively compared activPAL measurements with direct observation during assessments at discharge from the ICU or acute hospital in eight participants with a median age of 56 (1st–3rd quartile 48–65) years and an Acute Physiology and Chronic Health Evaluation II score of 23 (1st–3rd quartile 17–24). Frequency of sit-to-stand transitions; time spent standing, stepping, upright (standing and stepping), and sedentary (lying/sitting); and total steps were described; analysis was performed using Bland–Altman plots and calculating the absolute percent error. Results: All sit-to-stand transitions were accurately detected. The mean difference on the Bland–Altman plots suggested an overestimation of standing time with the activPAL of 31 (95% CI: −9, 71) seconds and underestimation of stepping time by 25 (95% CI: −47, −3) seconds. The largest median absolute percent errors were for standing time (21.9%) and stepping time (18.7%); time spent upright (1.7%) or sedentary (0.3%) was more accurately estimated. The activPAL underestimated total steps per session, achieving the largest percent error (70.8%). Conclusion: Because it underestimated step count, the activPAL likely incorrectly recorded stepping time as standing time, so that time spent upright was the measure of activity with the smallest error. Sedentary behaviour, including frequency of transitions, was validly assessed.