The LSF (least slack first) algorithm assigns a priority to a task according to its executing urgency. The smaller the remaining slack time of a task is, the sooner it needs to be executed. However, LSF may frequently cause switching or serious thrashing among tasks, which augments the overhead of a system and restricts its application. Assigning a preemption threshold in the scheduling policy can decrease the switching among tasks, however, the existing assigning methods are limited to the fixed priority such that they are not applied to the LSF algorithm. In order to relieve the thrashing caused by LSF, some applicable assigning schemes are presented to the LSF algorithm based on the preemption threshold. Every task is dynamically assigned a preemption threshold that is dynamically changing with the executing urgency of the task and is not limited by the number of tasks. Simulations show that, by using the improved LSF policy, the switching among tasks decreases greatly while the missed deadline percentage decreases. The proposed algorithm is useful for designing and implementing a real-time operating system.