Abstract：Due to inefficient use of traditional single-computing and disability to meet the instantaneity requirement with the increasing of radar tasks′ complexity, cluster computing is widely envisioned as a powerful paradigm for building useful distributed applications in parallel scheduling radar tasks. An improved allocation algorithm is proposed for radar task allocation of the distributed heterogeneous system. Based on the traditional allocation method, the continuous detection of task information of the processor is put forward. Furthermore, task accumulation threshold is defined to estimate each processor is overloaded or not. In particular, the task will be allocated to the processors which are detected in idle or light loading with the workload according to the threshold. Compared with the traditional method, simulation results on different processing performances and task accumulation thresholds show that the processing time of improved algorithm is reduced greatly and the load balancing ability is also improved immensely. What′s more, this algorithm works better in those systems where the processing ability among processors differs a lot.