Traditional monitoring methods of grinding wheel wear mainly take on-machine measurement systems or black-box algorithms. The first method has poor real-time performance for it can only be implemented after machining, making it difficult to adjust grinding processes in time. The second method has not revealed the mechanism of wear, resulting in poor model accuracy and flexibility. Besides, previous models are mainly for cylindrical wheels in surface grinding, while the properties of non-cylindrical wheel wear for gear grinding are different. To address this issue, a novel online monitoring method of non-cylindrical wheel wear for gear grinding is proposed. Firstly, a new parameter is defined based on grain height variation to describe the non-cylindrical wheel wear condition, avoiding the problem of underestimating the actual results due to a small measurement range. Then, a theoretical model for dynamic grinding forces is established considering more grain parameters and more complicated and realistic shape, so that on the basis of that model, the online monitoring method for wheel wear is proposed. Finally, experiments under different grinding conditions are conducted to verify the proposed models. The average calculation error of the grinding force model in different grinding parameters is 6.67%, and the maximum prediction error of the monitoring method under different wheel wear conditions is 8.2%.