<p>In this article, we develop a jackknife model averaging (JMA) method for modal linear regression with rightly censored responses. The weights in model averaging are obtained by maximizing the leave-one-out cross-validation criterion function involving the kernel density estimation, inverse-probability-of-censoring weighting and Kaplan-Meier estimation techniques. Under rather mild conditions, we establish the asymptotic optimality of the JMA estimator in the sense of minimizing out-of-sample prediction risk asymptotically. Besides, we also obtain some theoretical properties of regression coefficient within each misspecified candidate model, including the convergence rate and asymptotic normality. Some simulations are conducted to evaluate the performance of the proposed JMA approach against some competing model selection and model averaging methods, which show that the suggested JMA method is superior to its competitors. The proposed JMA method is also applied to a real data analysis as an illustration.</p>