In order to solve the problems of long simulation time and many parameters to be optimized in traditional antenna design methods, this paper proposes an antenna inverse modeling method based on the attention mechanism CNN-BiLSTM, which extracts the multidimensional features of the data through convolutional neural network (CNN), and captures the mapping relationship between the electromagnetic response and the geometric parameters of the microwave device by using bidirectional long and short-term memory network (BiLSTM). The introduction of the self-attention mechanism allows the model to notice the correlation between different inputs in the global, maximizing the retention of features and thus improving the robustness of the model. In this paper, the model is applied to the design process of printed dipole antenna. Through comparison, the inverse modeling method shows good prediction and generalization ability, which can be applied to the optimal design of antennas.