In order to evaluate the reliability of the cylindrical worm drive mechanism, the Taylor series expansion is used to expand the state function of the worm gear mechanism and the first four-order center moments of the structural state function are obtained. Regarding this as a constraint condition, the probability density functions of the worm gear tooth surface contact strength and tooth root bending strength structural state function are obtained by combining the maximum entropy principle. On the basis of the probability density functions, the reliability of the two failure modes of the worm gear is analyzed. On the other hand, considering the correlation between the two failure modes of worm gear tooth surface contact and tooth root bending, the reliability model under failure-related conditions is established, and the correlation coefficient between the two failure modes and the overall reliability of worm gear are given. Taking the worm gear of a reducer as an example, the proposed method is verified by Monte Carlo simulation. The results show that the reliability method based on the fourth moment and the maximum entropy principle has higher precision, and there is a certain correlation between the two failure modes of the worm wheel. Therefore, the failure-related worm wheel reliability assessment method is of great significance for the comprehensive reliability assessment of worm drive.