Real-time recognition of cutting tool states plays an important role to improve the efficiency and quality and reduce the cost of CNC machining. For complex parts with single-piece or small batch production, the geometry and cutting parameters in the machining process are constantly changing which makes a great challenge for accurate recognition of cutting tool state. To address the above problems, this paper presents a cutting tool state recognition approach based on cutting force signals, geometric information and process information. The cutting force signals of different tool states are collected and analyzed. The time analysis and time-frequency analysis methods are used to extract the features of cutting force signals. Associated with the machining process information and the geometric information of the parts, the input vector is established. Cutting tool state recognition model based on neural network is established and trained. The real-time recognition of cutting tool state is realized with the neural network model during the actual machining process. Experiments show that the approach can solve the real-time recognition of cutting tool states.