For the issues that the time efficiency problem of traditional Top-k join algorithm when dealing with massive data, a load-balanced parallel Top-k join query algorithm (P-TKJ) based on MapReduce framework is proposed. It used histograms to store data helps to increase CPU utilization. An early termination strategy and disk data selective access mechanism is incorporated to improve the performance of HDFS data access. In addition, a load balancing strategy based on the longest processing time-first (LPT) algorithm is proposed to balance the Reduce tasks, so as to design an efficient parallel Top-k join algorithm. A cluster experiment shows that this method can shorten the execution time of the algorithm effectively.