Dwell time is the amount of time required for performing boarding and alighting activities at stops. Under peak-load conditions, the dwell time can significantly increase due to a higher friction between on-board passengers when alighting and boarding. The influence of on-board crowding on increasing dwell time is indisputable. Herein, we develop a mixed-integer nonlinear programming (MINLP) model to optimize limited-stop patterns for bus services to minimize user and operator costs. The number of non-stop consecutive buses authorized to skip a station and in-vehicle crowding conditions are explicitly considered in our modeling framework. The benefits of limited-stop bus services are mainly overestimated in previous studies that ignore such operating conditions. Moreover, a genetic algorithm is developed to solve the problem in real-world cases. The findings show that the implementation of a limited-stop bus service can reduce in-vehicle travel times for passengers and operating costs for bus agencies in all-demand cases. Nonetheless, it can increase waiting times for users whose origin or destination stations are skipped due to the implementation of limited-stop services. Thus, the desirability of a limited-stop service can decrease with the growth of the demand level.