The invention discloses a tobacco cutting process parameter weight determining method based on random forest regression. The method using stable-state data sample in all-batch data during the tobacco cutting process, includes the steps of selecting research variables, sorting out interpretation variables through a Pearson correlation matrix, conducting modeling analysis through random forest regression, verifying the fitting effect and extrapolation performance by using goodness-of-fit and five-fold intersect verified standardized mean square error, and finally measuring the interpretation variable influence weight according to the average successively decreasing value according to OOB mean square error. The method overcomes the defects of subjective randomness of a subjective weight determining method and low prediction robustness of a major constituent weight determining method, has better noise tolerance and extrapolation performance, and has great advantages in processing high-frequency real-time data.