A tool magazine and automatic tool changer is used to store and change tools in a machining center. Its failures would increase the maintenance time and repair cost. For decreasing failures, an early warning system of failures can be used to detect characteristic signals which can reflect potential failures. The tool-pulling force is increasing to peak value instantly when the tool changing arm draws a tool holder out of a tool pocket. To predict failures, the tool-pulling force signals are analyzed and the relationship between the maximum tool-pulling force and the severity of the lock spring failure is established, then, calculation method of the tool-pulling force threshold of the lock spring failure which can lead to tool falling failures is developed and demonstrated. The verification testing results show that the failure warning method based on the tool-pulling force threshold is feasible to predict the tool falling failures.