To improve the real-time monitoring accuracy of the actual wear of cutting tools, Extreme Learning Machine (ELM) algorithm was applied to establish the tool wear model. An Improved Artificial Bee Colony (IABC) algorithm with a virtual bee was proposed to tune the input layer weights as well as the hidden layer biases in ELM. Time-domain analysis and Empirical Mode Decomposition (EMD) were used to extract the time-domain features and energy ratio of Intrinsic Mode Function (IMF) from cutting force signal, vibration signal and acoustic emission signal, and the features sensitive to tool wear were selected as the monitoring features. Then the tool wear value could be calculated through the trained model. The result show that the optimized ELM can predict the tool wear value accurately, and has a simpler network structure. Meanwhile, the IABC algorithm also shows a better searching ability.