Robots pose many benefits for manufacturing, including a high flexibility, low investment cost, and a small footprint compared to conventional milling machines. Their usability is limited because of a high likelihood of chatter and a comparatively low accuracy. Exact knowledge of the robot’s stiffness can be used to compensate for accuracy errors due to static deflection. The identification of the robot’s stiffness is typically a laborious process, as the stiffness model is usually not provided by the robot’s manufacturer. In this article, a method for identifying the stiffness of a milling robot using solely data from internal sensors is presented. The usage of a rotary dynamometer for force measurements and secondary encoders for joint deflection measurements is motivated by eliminating the need for costly external measurement equipment. Additionally, the data generation for the proposed method can be conducted during milling, reducing the downtime of the robot and enabling an in-process stiffness identification. Using up-to-date data also allows for detecting changes in the stiffness, e.g., due to wear. The identified local Cartesian stiffness can be used to optimize milling paths to reduce static deflection and therefore improve milling accuracy.