In this paper, a multimodal teaching model based on vector autoregressive (VAR) model is constructed for English teaching in colleges and universities, and the model is optimized and improved. The VAR model can predict the potential future paths of the specified variables in the model, and thus measure the long-term equilibrium relationship of multiple variables in English teaching in colleges and universities. This model was empirically studied in the multimodal teaching of four listening, speaking, reading and writing aspects. The results of the study show that the students in the experimental group significantly improved their performance in the listening test, in which the average scores of Groups A, B, and C were 18.49, 20.265, and 21.99, respectively, with a p-value of 0.000, which is lower than the 0.05 level of significance. Most of the students think that the multimodal English film and video clip teaching mode helps improve English thinking ability and fluency of oral expression. Regarding reading comprehension, the average score of the experimental class reaches 20.89 while the control class is 16.8, with a p-value of 0.001, indicating that the practical class is significantly better than the control class. Regarding writing instruction, most students also preferred multimodal instruction and recognized its positive effects. The application of multimodal teaching mode based on VAR model in English teaching in colleges and universities is practical and can significantly improve students’ listening, speaking, reading and writing abilities.