Falls are the single most dangerous safety accident within the construction industry, representing 33% of all fatalities in construction. Numerous unrecognized near-miss falls exist behind every major fall accident. The detection of near-miss fall occurrence therefore helps the identification of fall-prone workers/tasks and invisible jobsite hazards and thereby can prevent fall accidents. This paper presents and evaluates the feasibility of a threshold-based approach for detecting the near-miss falls of construction iron-worker. Kinematic data of subjects are collected through an IMU sensor attached to the subjects' sacrum; the subjects then perform walking on a steel beam structure. Fall-related features - sum vector magnitude (SVM), and normalized signal magnitude area (SMA) - are used to detect near-miss falls. Threshold values of these features are defined to achieve the best accuracy in near-miss fall detection based upon experiment data. According to selected threshold values, iron-workers' near-miss falls were detected. The result of this research demonstrate the opportunity of utilizing SVM and SMA in documenting workers' near-miss fall incidents in real-time.