Accidental falls (slips, trips, and falls from height) are the leading cause of death and injury on a construction site. Assessing the risk of such falls, therefore, becomes a fundamental step toward reducing these accidents. However, the quantitative assessment of a fall risk for construction workers is still very challenging because of sparse data related to fall accidents. Recently, there has been a growing interest in the identification of near-miss fall accidents to utilize them as supplementary data for fall-risk assessments. Current documentation for near-miss fall accidents is based on workers' self-reporting, a fact that adds variability to the data. In response, this research introduces a method that can detect near-miss fall incidents based on inertial measurement units (IMUs). A preliminary laboratory experiment collects data on ironworkers' typical movements, postures, and near-miss fall accidents. Workers' postures and movements are recognized through supervised classification algorithms; near-miss fall incidents during the classified posture/movement are quantifiably detected based on the time-series anomaly detection approach. Such research helps to identify the possibility of fall accidents more precisely according to worker's activity data. Additionally, documenting near-miss fall data provides quantitative data for ironworkers' fall-risk assessment, a significant step forward in the field.