For periodic tasks in a distributed real-timesystem, a number of static allocation algorithms have been developed which solve the problem of assigning and scheduling tasks effectively under some determined conditions. The principal limitation of these approaches is that the attributes of the tasks have to be known. Sometimes the execution time or the number of subtasks of a periodic task might be a stochastic process obeying some rule. In such cases, the performance of the static schemes will decrease greatly. According to the analysis of the processing in specific application fields, the authors model two types of random tasks in distributed real-time systems and introduce the static allocation algorithms (SAA) which have been applied in engineering for the two task models separately. On the basis of SAA, a predicting allocation algorithm (PAA) is presented for the assignment and the scheduling of multitasks in distributed systems. The proposed algorithm, depending on the statistic features of the task execution time or of the number of subtasks included in the tasks, can predict the task parameters reasonably and implement dynamic allocation of the tasks, so that the system can meet the timing requirements more efficiently. The results of the simulation of the two task models have shown that compared with SAA scheme, the performance of PAA is significantly better in task finishing time, load balancing, system response time, ratio of discarded tasks, etc.