ETL is the essential step to obtain high-quality data for data warehouse, and plays an important role in the construction and implementation of data warehouse. Aiming at the deficiency of traditional serial ETL process, we propose a parallel method for ETL based on agent and activity priority. This method first calculates the priority of each ETL activity and then utilizes the agent theory and multi-thread computing techniques to achieve parallel execution of independent ETL activities with the same priority. Experimental results show that this method achieves high speedup when the data volume is large and improves the efficiency of ETL process.