This study proposes a variable dynamic threshold for a jerk signal for achieving contact detection in industrial robots. A sensorless force control is useful for application to advanced industrial robots. The external torque of each joint is estimated accurately by using a disturbance observer and by inverse dynamics calculation. However, the phase-lag of the disturbance observer and the dynamic parameter error cause force estimation errors. These errors hamper quick and reliable contact detection in industrial robots. Therefore, this study proposes a useful contact detection approach that uses a variable dynamic threshold for a jerk signal. The proposed variable dynamic threshold reduces contact detection failures caused by force estimation errors during the free motion of a robot. The proposed threshold is suitable for robot motion and improves the speed of contact detection. Numerical simulation results show that the proposed method is valid for parameter variations in an actual industrial robot. In addition, experimental results show that the dynamic threshold is a useful technique even for application to general robot systems. The dynamic threshold is useful for application to any industrial sensorless force control system.