Subarray design is widely applied in large planar array monopulse system to synthesize optimal sum and difference patterns at a low system cost. Based on the excitation matching principle, subarray design can be formulated as a clustering problem which can be solved by K-means clustering method. When the elements are weighted uniformly, K-means clustering method can provide the optimal beam pattern synthesis performance. However, the performance will deteriorate when the elements are weighted non-uniformly. Therefore, a weighted K-means clustering method is proposed in this paper to improve the beam pattern synthesis performance for non-uniform element weights. The effectiveness of the proposed method is verified by numerical simulations.