Background: This study aims to perform a concurrent criterion validation of the activPAL3 activity monitor, in the detection of physical activity, steps, and postural transfers in older adults using video observation. Method: Twenty community-dwelling older adults performed both an unsupervised free-living activity protocol in their home environment, recorded using body-worn cameras, and a semi-structured supervised protocol in a smart-home setting, recorded using wall mounted cameras, with an activPAL3 attached to the thigh. Percentage of agreement and typical statistical accuracy metrics were calculated by comparing the activPAL3 output and the video observation gold-standard (0.04 s resolution). Results: The activPAL3 provided a valid measure of standing, sitting, lying, and purposeful walking, including stair climbing. Shuffling, picking, transition, and kneeling were not consistently classified when compared to video observation and were thus confounding activities for the activPAL3. Sedentary behavior was better identified in a free-living scenario than during the semi-structured protocol. Step detection during stair ascending and descending achieved a high percentage of agreement (>89%). Steps detected during walking were underreported (80.2% free-living, 72.9% laboratory-based). Many steps were not detected during shuffling and transitions; overall, the percentage of agreement was low (59.5% free-living, 58% laboratory-based). Good sensitivity, specificity, and accuracy (>85%) were achieved for laboratory-based activities and good to excellent sensitivity, specificity, and accuracy (>89%) were achieved for free-living activities. Percentage of agreement was higher for free-living activities (85.2%) compared to laboratory-based activities (69.15%). Conclusion: This validation study provided a detailed insight into the physical activities that the activPAL3 classifies in its three main activity categories, step detection and postural transition analysis in a laboratory and a free-living setting. Caution is advised when measuring relatively more intensive physical activity protocols (e.g., in-lab), assessing postural transfer quantity, or during sedentary behavior analysis, as some short-duration sedentary bouts are ignored and postural transfers underreported.